Bioinformatics analysis of Rickettsia typhi autoimmune associations and screening of Streptomyces-derived inhibitors Authors: Zarrin Basharat, Hanan A Ogaly, Fatimah A M Al-Zahrani, Calvin R Wei, Syed Shah Hassan, Muhammad Salman, Madiha Islam, Ibrar Ahmed, Yasir Waheed, Seil Kim Published: 2025 ABSTRACT AbstractBackgroundRickettsia typhi is the causative agent of epidemic murine typhus and Rocky Mountain spotted fever. The infection can affect multiple vital organs, including the heart, lungs, kidneys, and brain. Doxycycline is the recommended treatment but inflammation, mal-response, and drug resistance may arise. No natural product inhibitors have been reported against this bacterium.AimThe objective of this study was to establish a potential connection between autoimmune disorders triggered by R. typhi, identify therapeutic targets within its core proteome, and explore novel natural product inhibitors from Streptomyces spp. that could potentially inhibit it.MethodologyComplete proteomes of four publicly available R. typhi strains were used for pan-proteomic analysis. The fni gene product (Isopentenyl pyrophosphate isomerase) was selected as the potential drug target. Molecular docking of 607 Streptomyces-derived metabolites was performed, with top hits validated using DiffDock and Vinardo scoring. Additionally, the Absorption, Distribution, Metabolism, Excretion, and Toxicity properties of the leading compounds were assessed via pkCSM, and formulation characteristics optimized using FormulationAI.ResultsOut of the 803 core proteins, associations between 14 proteins were mined for autoimmune diseases (including psoriasis, rheumatoid arthritis, optic atrophy, uveitis, even-plus syndrome, Sjogren syndrome, inflammatory bowel disease, allergic rhinitis, systemic lupus erythematosus, sclerosis, Stevens-Johnson syndrome, toxic epidermal necrolysis, colitis etc.). 17 core proteins were predicted as druggable. ZINC01482946 demonstrated the strongest inhibitory potential, as confirmed by DiffDock scoring, convolutional neural network-based ranking, and Vinardo scoring. It demonstrated a stable configuration and exhibited a favorable pharmacokinetic profile, with bioavailability enhanced through cyclodextrin complexation.ConclusionTo the best of our knowledge, this is the first report identifying human autoimmune associations with R. typhi and natural product inhibitors targeting the pathogen. ZINC01482946 shows potential as an effective inhibitor of R. typhi, while SBE-β-CD appears to be a promising cyclodextrin for improving its solubility and bioavailability.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13040-025-00499-w.Keywords: Rickettsia typhi, Rocky mountain fever, Streptomycetes, Computer aided drug design, Inhibitor, Bioinformatics Introduction Rickettsia typhi is a gram-negative, obligatory intracellular bacterium that causes endemic typhus [1, 2]. It moves using actin-based motility, often in a circular motion with limited directional control [3–5]. R. typhi is difficult to isolate and cannot be cultured using standard plating techniques used with the majority of other species [6]. The bacterium infects endothelial cells, causing systemic vasculitis that can affect multiple organs. Clinically, infection presents with high-grade fever, headache, nausea, myalgia, fatigue, and rash. It can also cause thrombocytopenia. Although there is a low death rate for this illness, it can impact key systems, including the brain, lungs, liver, kidneys, and heart, which in more severe cases can result in multisystem disease like meningoencephalitis, interstitial pneumonia, as well as respiratory failure [7, 8]. R. typhi is maintained in a zoonotic cycle primarily involving rodents as reservoirs and the Oriental rat flea as the principal vector [9], with human infection occurring through inoculation of infected flea feces into skin abrasions or mucous membranes, and less commonly via aerosol or conjunctival exposure [10, 11]. After introduction into the host, R. typhi infects endothelial cells, leading to systemic vasculitis and multi-organ involvement [12]. Mortality rates of murine typhus sickness are around 5% [13] but rates as high as 30–70% have been reported from Indian studies [14]. Elderly patients who frequently need to be hospitalized as the disease worsens require extra attention when it comes to proper murine typhus management. Outbreaks of murine typhus have been reported across the world, mostly in tropical and subtropical regions. Recent reports have come from Mexico, Europe, the Middle East, Africa, and Asia in addition to America and Australia [15]. Prevention primarily relies on avoiding exposure to the causative organism. Vaccination remains an essential strategy for effective disease control but currently no vaccines are available at the commercial level that target any molecular component of the pathogen [16–18]. There are currently a limited number of medications available for the treatment of Rickettsia infections. Antibiotics such as tetracyclines, (e.g. doxycycline), macrolides, or chloramphenicol are effective against Rickettsiae sp. due to their ability to inhibit bacterial protein synthesis at the ribosomal level. Rifampin, which inhibits bacterial RNA polymerase and thereby blocks RNA transcription, is also used in some cases. To date, doxycycline remains the drug of choice for treating all Rickettsia infections [10, 19]. Recent clinical trials have reinforced the superior efficacy of doxycycline compared to alternatives like azithromycin in treating R. typhi infections, with azithromycin showing higher rates of treatment failure and prolonged fever clearance times [20]. Chloramphenicol remains a secondary option but is associated with higher case fatality rates and adverse effects, including bone marrow suppression [21] and aplastic anemia [22]. Its use is further limited in pregnant women due to potential risks such as teratogenicity and the development of gray baby syndrome in neonates [23, 24]. Fluoroquinolones are generally discouraged due to poor clinical outcomes, emerging resistance, and a lack of strong evidence supporting their efficacy against Rickettsia infections [25]. Some patients exhibit poor responsiveness to doxycycline, and alternative options remain limited. While rifampin is sometimes used, its efficacy against epidemic typhus caused by R. typhi remains uncertain [10, 19]. Moreover, increasing reports of antibiotic resistance in endemic regions have raised concerns that R. typhi may eventually develop resistance to doxycycline [26, 27]. Persistent Rickettsia infections, including those caused by R. typhi and related species, have been documented even after antibiotic treatment [28, 29]. Given the absence of a licensed vaccine and the growing threat of antibiotic resistance, discovering novel molecules for effective intervention against R. typhi is necessary. In this study, we screened the Streptomyces spp. metabolites to identify potential inhibitors targeting R. typhi. Research into autoimmune disease association and molecular mimicry is crucial for unraveling the complex interplay between the immune system and pathogens. Autoimmune diseases arise when the immune system mistakenly attacks body tissues and prompts investigations into genetic predisposition and environmental triggers. Rickettsia spp. have been implicated in autoimmune disorders. Paiardini and Pascarella have previously used bioinformatics strategy and predicted molecular mimicry between the immunodominant regions of human soluble liver antigen/liver-pancreas (SLA/LP) and a surface antigen (PS 120) from Rickettsia spp., causing autoimmune hepatitis [30]. Marangou et al. [31] documented the association between human thyrotropin and the ATPase of R. prowazekii, resulting in Graves’ disease, which is a molecular mimicry-induced autoimmune disorder. R. typhi has also demonstrated the capacity for long-term persistence in animal models, including in immunocompetent mice where it was recoverable from brain, spleen, and lungs up to a year post-infection [32]. Its reactivation in the CNS of immunodeficient mice led to neuroinflammation and neuron loss, suggesting potential for chronic immune activation. Human studies further reveal significant serological cross-reactivity between Rickettsia species, complicating diagnosis and raising the possibility of shared antigens [33]. Elevated anti-Rickettsial IgG titers have been observed in patients with autoimmune diseases such as SLE and ITP, though it remains unclear whether this reflects prior infection, immune dysregulation, or cross-reactivity of autoantibodies [34]. These findings collectively support the hypothesis that R. typhi may contribute to autoimmune pathogenesis through chronic antigenic stimulation, molecular mimicry, or immune-mediated tissue damage. While previous studies on other Rickettsia spp. have suggested that molecular mimicry may contribute to autoimmune-like responses and chronic inflammation, the specific role of mimicry in R. typhi remains underexplored. We hypothesized that the R. typhi proteome harbors previously unexplored molecular mimicry with human proteins, which may contribute to autoimmune responses or facilitate immune evasion. To capture a comprehensive view of this phenomenon, we employed rapid computational data mining from available databases. Since R. typhi is the causative agent of endemic typhus, a disease with significant morbidity and complex immune interactions, understanding molecular mimicry could elucidate how the bacterium avoids immune clearance or triggers host immune dysregulation. This knowledge is essential not only for deciphering disease progression but also for identifying novel therapeutic targets. By pinpointing host protein regions that resemble bacterial antigens, it may be possible to develop interventions that disrupt harmful cross-reactive immune responses or design vaccines that avoid autoimmune sequelae. Thus, studying molecular mimicry in R. typhi bridges a critical gap between its molecular pathogenesis and, aligning with broader efforts to combat rickettsial diseases through targeted immunomodulation and precise drug design. In parallel, we proposed that metabolites from Streptomyces spp., a rich source of antibiotics, could serve as potential inhibitors of R. typhi, offering a promising therapeutic avenue. Accordingly, we analyzed homology between R. typhi and human proteins to identify sequence mimics and assess their potential role in triggering autoimmune disorders. Such mimics may play a pivotal role in evading immune responses or modulating host cell functions. We also screened potential inhibitors from a Streptomycete library and validated their efficacy through molecular dynamics (MD) simulations and Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) analysis. The insights gained from this study deepen our understanding of the mechanisms by which R. typhi interacts with and adapts to the host immune system as well as how it may be inhibited using natural products. Materials and methods Data retrieval Bacterial and Viral Bioinformatics Resource Center (BV-BRC) [35] was used to retrieve the complete proteome sequences of R. typhi isolates. The strains were: B9991CWPP (GenBank accession: CP003398.1), TH1527 (GenBank accession: CP003397.1), TM2540 (GenBank accession: LS992663), and Wilmington (GenBank accession: AE017197) (retrieved on 1st May 2023). The human proteome was retrieved from the Universal Protein Resource (UniProt ID: UP000005640; Total sequences = 83,587) database to analyze non-homologous proteins [36]. The overall schematic of the workflow is depicted in Fig. 1. Schematic overview of the major assays conducted in this study. In the venn diagram depicting pan-proteome, exclamation marks (!) indicate unique proteins; ‘C’ denotes core proteins; asterisks (*) represent shared accessory proteins Pan-proteomics of R. typhi Bacterial Pan Genome Analysis tool (BGPA) was used to perform pan-proteomic analysis to examine the similarities and differences identified in proteome fractions [37]. The USEARCH clustering algorithm was used to determine the orthologous proteins in the pan-proteome by using a 70% cut-off value instead of 50%, to ensure a more reliable analysis. Previously Thorpe et al. reported similar threshold (≥ 70% identity with ≥ 50% query coverage) to define putative 1:1 orthologues, supporting the use of this stringent criterion [38]. The MUltiple Sequence Comparison by Log- Expectation (MUSCLE) program [39] was used for proteome alignment, and phylogenetic analysis was done by unweighted pair group method with arithmetic mean (UPGMA) method. The functional annotation of the pan-proteome was accomplished by the clustering of ortholog groupings to identify functional groups of the homologous proteins [40]. Molecular mimicry and disease association mapping Disease-associated pathways of R. typhi were investigated through molecular mimicry mapping. Homologous proteins of R. typhi were compared with the human proteome using stand-alone Basic Local Alignment Search Tool (BLAST) (identity threshold ≥ 50%; bit score: ≥100-bit score to retain proteins for further analysis) [41, 42]. This was carried out across accessory, core, and unique fractions of the proteome. Identical sequences of ≥ 10 amino acids were classified as molecular mimics, highlighting peptides that may interact with the host immune system and modulate cellular functions. Such mimics not only provide insights into pathogen-host interactions and potential autoimmune responses but also serve as candidates for drug screening and structural analyses. For each homologous protein, human protein was mapped to its UNIPROT protein ID and searched in PHAROS database (https://pharos.nih.gov/diseases; retrieved 12 December 2023) [43] for associated disease pathway. PathDIP database [44] was used for predicting significant associations between R. typhi proteins and curated human disease pathways to enrich the pathway information. Therapeutic candidate prediction and prioritization For drug design, non-homologous essential proteins were inferred through subtractive proteomics approach. Two databases were used for mining the essential proteins to evaluate essential proteins. i.e., Cluster of Essential Genes (CEG) [45]and Database of Essential genes (DEG) [46]. The union of proteins from both databases was considered because the CEG and DEG analyses identify essential protein sets from complementary perspectives. CEG analysis captures genes conserved across species that are indispensable for baseline cellular processes, representing the core housekeeping machinery. In contrast, DEG analysis highlights proteins whose essentiality is supported by experimental data, often reflecting condition-specific dependence under environmental or pathogenic contexts. The observed overlap represents proteins that are both evolutionarily indispensable and experimentally validated, strengthening their candidacy as robust targets and ensuring downstream prioritization is based on universally essential as well as contextually relevant proteins. We also conducted a comprehensive pathway mapping and functional categorization using Kyoto Encyclopedia of Genes and Genomes (KEGG) BlastKoala [47] and KEGG Mapper [48] for these proteins. Potential therapeutic targets should be non-homologous to humans to avoid drug-molecule cross-reactivity. To accomplish this, the acquired number of proteins were translated to proteins and then used for identifying non-homologs against human proteome, using BLASTp (E-value = 10− 2) [49]. For prediction, 70% similarity, 50% query sequence coverage, a gap penalty of 11 and a gap extension penalty of 1 were utilized to balance sensitivity and specificity in homology detection. Pathway mapping was done and compared to essential protein list. As a further step to ensure safety of the predicted drug targets, the identified proteins were also screened against proteins from the human gut microbiota to eliminate targets that share homology with beneficial gut flora [50]. This approach helps to reduce the potential for disrupting the gut microbial community when targeting pathogenic proteins. The same BLASTp parameters were used for gut microbiota screening, and hits were evaluated to exclude homologs to gut flora proteins, supporting the specificity of targets selected. The latest version (2021-3) of the DrugBank database [51]was used to determine the druggability capacity of the hit proteins [52]. The selected target was further validated through a BLAST search (https://www.genome.jp/tools/blast/; scoring matrix: BLOSUM62) in the KEGG pathway database [53] to confirm its functional annotation. Virtual screening 3D structure of the selected target Isopentenyl pyrophosphate (IPP) isomerase (also called Isopentenyl-diphosphate delta-isomerase), Accession no: WP_011190900.1, was retrieved from Alpha fold database (https://alphafold.com/entry/Q68WS6) [54]. The target protein did not share homology with human IPP isomerase, supporting its potential as a selective therapeutic target. The next step involved ligand-based virtual screening. The Streptomyces-derived metabolites were selected from a curated subset of the ZINC12 database (https://zinc12.docking.org/catalogs/streptome; retrieved 18 February, 2020) for screening. The selection process involved multiple filtering criteria like drug-likeness (Lipinski’s rule of five), optimization (Oprea’s lead-likeness) [55] and non-mutagenicity. The successive filters were applied to retain compounds with favorable pharmacokinetic and toxicological profiles. The resulting set of 607 Streptomyces metabolites represented a chemically diverse group of bioactive compounds with potential therapeutic relevance. Streptomyces spp. metabolites are widely recognized as a rich and diverse source of bioactive compounds, including many clinically important antibiotics [56]. The genus is responsible for producing approximately 70–80% of naturally derived antibiotics used in medicine, showcasing remarkable structural diversity and pharmacological activities [57]. The secondary metabolites serve various biological roles such as antimicrobial, antiviral, anticancer, immunosuppressive, and enzyme inhibitory functions [58]. Owing to their biochemical diversity and therapeutic relevance, these metabolites were selected as a promising source for identifying novel inhibitors against R. typhi. Prior to screening, all library compounds were energy minimized and protonated at physiological pH (7.4) to prepare them for virtual docking. Screening was conducted using AutoDock Vina [59], after converting the protein to pdbqt format and adding hydrogen atoms at pH 7.4. The grid parameters were: Grid center: X = 1.71, Y = 1.91, Z=-1.25; Grid size : X = 14.64, Y = 11.86, Z = 14.76; Grid space: 0.375. The top conformations of the compounds with the lowest binding affinity values were chosen for further analysis. AutoDock Vina is a popular classical molecular docking tool. It utilizes an empirical scoring function that combines steric (van der Waals-like) interactions, hydrophobic effects, a non-directional hydrogen bonding term, and penalties for conformational entropy [59]. A global optimization algorithm is employed to predict ligand binding poses and estimate binding affinities. To validate findings from AutoDock Vina, machine learning (ML) methods were employed. Integrating AutoDock Vina with ML approaches leverages their complementary strengths because Vina rapidly generates plausible conformations, while ML-driven scoring refines pose selection by identifying subtle interaction patterns learned from large protein-ligand datasets. DiffDock [60], a state-of-the-art docking method based on diffusion generative models, generates ligand poses by learning the distribution of possible binding conformations, rather than relying on traditional scoring optimization. This model has demonstrated superior speed and accuracy, reframing molecular docking as a generative modeling task. Additionally, convolutional neural network (CNN)-based scoring functions, such as those implemented in GNINA [61] are trained on large collections of experimentally resolved protein-ligand complexes. These models learn interaction features directly from 3D molecular structures, capturing non-linear and context-specific patterns that classical scoring functions often miss. Using docking module at Bacterial and Viral Bioinformatics Resource Center (BV-BRC) server [35], DiffDock confidence scores were obtained for priority ligands to evaluate how well each ligand fit the binding pocket. These fitting poses were then assessed using GNINA, which performed CNN-based scoring and affinity prediction. CNN scoring evaluated docking potential based on ligand shape, volume, and complementarity, while CNN-based affinity prediction estimated binding strength by considering chemical bonds, hydrophobic and electrostatic interactions, and the conformational flexibility of both ligand and receptor. Protein-ligand interaction energy was evaluated by Vinardo score [62]. Molecular dynamics (MD) simulation MD simulation was carried out using Groningen MAchine for Chemical Simulation (GROMACS) version 2021 [63]. Protein and ligand topologies were prepared by adding missing atoms and assigning Optimized Potentials for Liquid Simulations- All Atoms (OPLS-AA) force field parameters. The solvated system was built using TIP3P water, and counter ions were added to neutralize the overall charge [52]. Energy minimization was performed to relax steric clashes, followed by stepwise equilibration under constant temperature and pressure with position restraints applied to biomolecules. The equilibration was done in two stages: constant volume (NVT) equilibration to stabilize temperature, followed by constant pressure (NPT) equilibration to stabilize system density and pressure. The Linear Constraint Solver (LINCS) algorithm was applied to constrain covalent bond lengths, ensuring stable integration of equations of motion. Long-range electrostatics were treated using the Particle Mesh Ewald (PME) method. Once equilibrated, production MD simulations were run for 100 ns. Root mean square deviation (RMSD) was used to measure the average change in displacement of protein-ligand complex atoms for over 100 ns with respect to the first frame. Root mean square fluctuation (RMSF) was determined to quantify residue-level flexibility of the protein [64]. Additional structural properties, including the radius of gyration (Rg), were computed to evaluate overall compactness and folding stability. Hydrogen bonding analysis was performed to monitor the persistence and number of intermolecular interactions between the protein and ligands throughout the trajectory. Time-resolved principal component analysis (PCA) was performed to explore collective motions and the first two principal components (PC1 and PC2). Furthermore, solvent-accessible surface area (SASA) values were extracted as time-dependent profiles to examine exposure of the complexes to solvent. All data were exported as Excel sheets from GROMACS analysis outputs, and subsequent visualization and comparative plotting were performed in Python (using libraries like pandas, matplotlib, seaborn). Thermodynamic changes were measured by change in various energy values. MM/GBSA function was used to extract ligand binding free energy. ADMET profiling Computational pharmacology and pharmacokinetic property analyses were performed. These comprised absorption, distribution, metabolism, skin penetration, excretion (ADME), and drug-like behavior of molecules using pkCSM program [65]. The toxicity profiles for mutagenicity, neurotoxicity, and carcinogenicity were also determined [66]. Physiological based pharmacokinetic (PBPK) parameters like bioavailability, plasma concentration etc., were also determined as described previously [67]. Formulation characteristics were improved using FormulationAI (https://formulationai.computpharm.org/; accessed 8 January 2025). Results Pan-proteomics It is inferred that a single proteome sequence cannot properly explain the whole proteomic repertoire of an organism [68]. Hence, the concept of pan-proteome is applied to explore the genetic diversity of species. It makes use of the whole protein pool from every strain belonging to a clade, genus, or species. The proteins are categorized into core proteome (the proteins present in all strains), accessory proteome (these proteins are in present two or more strains), and unique proteome (these proteins are found just in a specific strain) [69]. The power-law regression model was used for pan-proteome prediction, where Ypan = Apan. xBpan + Cpan. Values of A, B, and C were 1,237, 0.334, and 1075.12, respectively. As a result, the pan-proteome of R. typhi was found as open (b value = 0.33). Due to continual evolution as well as horizontal gene transfer, there is a hypothesis that there is a property of pan-proteome that will be always open for the bacterial species [70]. There were more than 23,000 DNA coding sequences in the pan-proteome of R. typhi (Supplementary Fig. 1). 1,307 proteins were strain-specific, compared to 21,594 accessory proteins. A high number of additional proteins were found in strain TA763 (n = 1162), while the number was low in strain sido (n = 335). The strain Karp contained the highest number of unique proteins (n = 181), whereas the lowest number of unique proteins were found in strain AFSC7 (n = 3). The highest number of missing proteins was (n = 151) recorded in the sido strain. A total of 803 of these proteins were identified as core and shared by all strains even after paralogs were removed. Core proteome-based drug design has been implemented previously [71] and hence, used for this species. The core proteome of R. typhi was found to be enriched in proteins involved in translation, ribosomal/protein biogenesis, repair, replication, recombination, and energy production/conversion, according to the Clusters of Orthologous Groups (COG) functional annotation for the pan-proteome of R. typhi. Proteins involved in repair, glucose metabolism, and general activities were found to be particularly abundant in the unique proteome (Supplementary Fig. 1). Autoimmune disease association and molecular mimicry study Core proteome showed 34 homologous proteins (Table 1) with host (> 50% sequence similarity), while accessory proteome showed only one gene product (hypothetical protein) similar to host Succinate dehydrogenase assembly factor 4 (Uniprot ID: Q5VUM1) with 57% similarity. However, no mimic (sequence length of ≥ 10 amino acids) was identified with this protein. The unique proteome fraction of R. typhi did not map to any human homolog. Table 1List of human homologs and related autoimmune disorders. Molecular mimics in R. typhi homologs have also been shownSerial no.Uniprot ID (name)Gene symbolAutoimmune disease mapped with PHAROSMolecular mimics in R. typhiBacterial homolog1.P21399 (Cytoplasmic aconitate hydratase)ACO1-YPDSLVGTDSHTTM,LGWGVGGIEAEA, VVGKFVEFFG, VVIAAITSCTNTSNPSVM, VLSGNRNFEGR, FNSYGSRRGN, GSSRDWAAKGP, ERIHRSNLVGMGAconitate hydratase AcnA2.Q96TA2 (ATP-dependent zinc metalloprotease YME1L1)YME1L1psoriasis, rheumatoid arthritis, optic atrophyGPPGTGKTLLA, EMFVGVGASRATP-dependent zinc metalloprotease FtsH3.P38646 (Mitochondrial, stress 70 protein)HSPA9autoimmune disease, uveitis, even-plus syndromeGIDLGTTNSCVAVMEGK, PAKRQAVTNP, KMKETAENYLG, AVITVPAYFND, INEPTAAALAYG, IAVYDLGGGTFD, GVFEVKSTNGDTFLGGEDFD, LVGGMTRMPKVQ, LLLDVTPLSLGIETLGGVFT, GIPPAPRGVPQIEVTFDIDANGIVHVSAKDKMolecular chaperone DnaK4.P31040 (Succinate dehydrogenase [ubiquinone] flavoprotein subunit, mitochondrial)SDHA-VVVGAGGAGLR, FPTRSHTVAAQGGI, YDTVKGSDWLGDQDAI, PFSRTEDGKIYQR, EFVQFHPTGIY, GERFMERYAP, AKDLASRDVVSR, PIPVLPTVHYNMGGIPTNY, NSLLDLVVFGR, ARKESRGAHAREDYSuccinate dehydrogenase flavoprotein subunit5.P10809 (60 KDa HSP; mitochondrial)HSPD1-TKDGVTVAKSI, AGDGTTTATVLARChaperonin GroEL6.P05166 (Propionyl-CoA carboxylase beta chain, mitochondrial)PCCBSjogren syndromeNDSGGARIQEGV, LIMGPCAGGAVYSPALTDF, VDVPGFLPGT, TVITRKAYGGAY,Acyl-CoA carboxylase subunit beta7.P25705 (ATP synthase subunit alpha, mitochondrial)ATP5F1A-PIGRGQRELIIGDRQTGKT, IAIDTIINQK, YCIYVAIGQKRS, CSMGEYFRDNG, HALIIYDDLSK, SLLLRRPPGREAYPGDVFYLHSRLL ERAAKM, IETQAGDVSAYIPTNVISITDGQIFLE, SVSRVGSAAQF0F1 ATP synthase subunit alpha8.P06576 (ATP synthase subunit beta, mitochondrialATP5F1B-ILVTGIKVVDLLAPY, KGGKIGLFGGAGVGKTVLIMELINNVAKAHGGY, VFAGVGERTREGNDLYHEMI, SKVALVYGQMNEPPGARARVAL, DNIFRFTQAG, EVSALLGRIPSAVGYQPTLATDMG, GSITSVQAIYVPADDLTDPAP, FAHLDATTVLSR, IAELGIYPAVDPLDS, YKSLQDIIAILGMDELSEEDKLTVSRARKIQRFLSQPF,F0F1 ATP synthase subunit beta9.P07954 (Fumarate hydratase, mitochondrial)FHPsoriasis, ulcerative colitisFPLVVWQTGSGTQTNMN, PVHPNDHVNK, SLMKIANDIR, LPENEPGSSIMPGKVNPTQClass II fumarate hydratase10.P09622 (Dihydrolipoyl dehydrogenase, mitochondrial)DLD-GCIPSKALLN, CHAHPTLSEADihydrolipoyl dehydrogenase11.O76031 (ATP-dependent Clp protease ATP-binding subunit clpX-like, mitochondrial)CLPX-GPTGSGKTLLAQTLAK, KTGARGLRSIATP-dependent Clp protease ATP-binding subunit ClpX12.G3V0I5 (NADH dehydrogenase [ubiquinone] flavoprotein 1, mitochondrial)NDUFV1PsoriasisSGLRGRGGAGF, YIYIRGEFYNEASN, GLIGKNACGSG, RGAGAYICGEETAL, HESCGQCTPCREG, IEGHTICALGDNADH-quinone oxidoreductase subunit NuoF13.Q9Y697 (Cysteine desulfurase)NFS1-EIIFTSGATESNN, SISGHKIYGPKG, GGGQERGMRSGT, SSGSACTSASLEPSYVLR, SIRFGIGRFTTEIscS subfamily cysteine desulfurase14.O75306 (NADH dehydrogenase [ubiquinone] iron-sulfur protein 2, mitochondrial)NDUS2allergic rhinitisDPHIGLLHRGTEKLIE, PYFDRLDYVS, IHHFKLYTEGY, EAPKGEFGVYLNADH dehydrogenase (ubiquinone) subunit D15.P00156 (Cytochrome b)MT-CYB-MATAFMGYVLPWGQMSCytochrome b N-terminal domain-containing protein16.P36957 (Dihydrolipoyllysine-residue succinyltransferase component of 2-oxoglutarate dehydrogenase complex, mitochondrial)ODO2psoriasisLTTFNEIDMS2-oxoglutarate dehydrogenase complex dihydrolipoyllysine-residue succinyltransferase17.P49411 (Elongation factor Tu, mitochondrial)EFTUinflammatory bowel diseaseGTIGHVDHGKT, DCPGHADYVKNMITG, DGPMPQTREHElongation factor Tu18.Q00266 (S-adenosylmethionine synthase isoform type-1)MAT1Apsoriasis, demyelinating disease-Methionine adenosyltransferase19.A0A7P0TAW3 (Transitional endoplasmic reticulum ATPase)VCPPsoriasis, Stevens-Johnson syndrome, toxic epidermal necrolysis-Holliday junction branch migration DNA helicase RuvB20.P03886 (NADH-ubiquinone oxidoreductase chain 1)MT-ND1--NADH-quinone oxidoreductase subunit NuoH21.P11177 (Pyruvate dehydrogenase E1 component subunit beta, mitochondrial)PDHBsystemic lupus erythematosus, systemic sclerosisPIVFRGPNGAPyruvate dehydrogenase complex E1 component subunit beta22.Q9UJZ1 (Stomatin-like protein 2, mitochondrial)STOML2Sjogren syndromeTQLAQTTMRSEStomatin-like protein23.P53597 (Succinate–CoA ligase [ADP/GDP-forming] subunit alpha, mitochondrial)SUCLG1-KTRLIGPNCPGVI, PGECKIGIMPGHIHK, IGIVSRSGTLTYEAV, IGEIGGNAEESuccinate–CoA ligase subunit alpha24.P21912 (Succinate dehydrogenase [ubiquinone] iron-sulfur subunit, mitochondrial)SDHBSjogren syndrome, rheumatoid arthritisGPMVLDALIKIKNE, DSTLTFRRSCREGICGSCAMNI, REKLDGLYECILCACCSTSCPSYWWN, LYRCHTIMNCTSuccinate dehydrogenase iron-sulfur subunit25.Q16740 (ATP-dependent Clp protease proteolytic subunit, mitochondrial)CLPPPsoriasisRGERAYDIYSRLL, MYINSPGGVVTAGLAIYDTMQ, SRIMIHQPSGGATP-dependent Clp endopeptidase proteolytic subunit ClpP26.Q9UMS0 (NFU1 iron-sulfur cluster scaffold homolog, mitochondrial)NFU1--NifU family protein27.O75251 (NADH dehydrogenase [ubiquinone] iron-sulfur protein 7, mitochondrial)NDUFS7Sjogren syndromeSLWPMTFGLACCAVEMM, TNKMAPALRKVYDQM, SMGSCANGGGYYH, SYSVVRGCDRIVPVD, PGCPPTAEALNADH-quinone oxidoreductase subunit B28.E7EPT4 (NADH dehydrogenase [ubiquinone] flavoprotein 2, mitochondrial)NDUFV2-PVGKYHIQVCTTTPC, ECLGACVNAPNADH-quinone oxidoreductase subunit NuoE29.O00217 (NADH dehydrogenase [ubiquinone] iron-sulfur protein 8, mitochondrial)NDUFS8Sjogren syndromeGEERCIACKLCEA, DGSRRTTRYDIDMTKCIYCG, CQEACPVDAIVEGPNFEFNADH-quinone oxidoreductase subunit NuoI30.Q9H1K1 (Iron-sulfur cluster assembly enzyme ISCU)ISCU-DHYENPRNVGSLDK, DHYENPRNVGSLDK, FKTFGCGSAIASSSLFe-S cluster assembly scaffold IscU31.P10109 (Adrenodoxin, mitochondrial)FDX1Ulcerative colitis-ferredoxin family 2Fe-2 S iron-sulfur cluster binding protein32.Q86SX6 (Glutaredoxin-related protein 5, mitochondrial)GLRX5Sjogren syndrome-Grx4 family monothiol glutaredoxin33.Q9P0M9 (Large ribosomal subunit protein bL27m)MRPL27--50 S ribosomal protein L2734.O75394 (Large ribosomal subunit protein bL33m)MRPL33--50 S ribosomal protein L33 List of human homologs and related autoimmune disorders. Molecular mimics in R. typhi homologs have also been shown YPDSLVGTDSHTTM, LGWGVGGIEAEA, VVGKFVEFFG, VVIAAITSCTNTSNPSVM, VLSGNRNFEGR, FNSYGSRRGN, GSSRDWAAKGP, ERIHRSNLVGMG The molecular mimics for R. typhi core proteins having homology with humans were mapped by alignment. Among these, 19 protein showed autoimmune disease potential (Supplementary Fig. 2). Eight of the proteins did not exhibit any mimic of ≥ 10 amino acids, whereas 26 proteins displayed mimicry potential. Succinate dehydrogenase flavoprotein subunit exhibited the highest number of molecular mimics (Supplementary Fig. 3). Among the proteins exhibiting mimicry, only 14 showed linkage to autoimmune disorders based on PHAROS database mapping (Fig. 2). The autoimmune diseases associated with them were diverse and included psoriasis, rheumatoid arthritis, optic atrophy, uveitis, even-plus syndrome, Sjogren syndrome, inflammatory bowel disease, allergic rhinitis, systemic lupus erythematosus, systemic sclerosis, and ulcerative colitis. Sankey plot depicting molecular mimics of R. typhi proteins and their possible association with autoimmune diseases. The mention of autoimmune disease indicates a correlation derived from the PHAROS database, reflecting potential immune-related associations rather than confirmed disease-specific identities Prediction of therapeutic targets To identify potential therapeutic targets, we first focused on essential proteins required for the survival of R. typhi from the core proteome. Since essential proteins are highly conserved throughout evolution and are consistently present across all strains of the pathogen, they represent attractive candidates for drug targeting [72]. A differential analysis of 803 core protein-coding sequences was evaluated using two specialized databases, DEG and the CEG. DEG identifies essential proteins through homology-based searches against a reference dataset of ~ 30,000 proteins from 70 species. In contrast, CEG predicts essentiality by organizing proteins into pre-defined homology-dependent clusters, thereby enhancing specificity and enabling visualization of conserved functionalities. Using this approach, CEG identified 437 proteins as crucial for R. typhi survival. The integration of the DEG and CEG datasets yielded 426 shared essential proteins that were used for further analysis (Supplementary Table 1). The substantial overlap between datasets underscores a strong correlation between the two approaches, with the shared set representing proteins that are both indispensable for core functions and supported by experimental validation, thereby reinforcing their potential as promising drug targets. A comprehensive pathway mapping and functional categorization of these 426 essential proteins was carried out and pathway reconstruction revealed a strong dominance of metabolic functional pathways (Supplementary Table 1). Several of the proteins were part of multiple pathways. Core pathways such as general metabolic pathways, biosynthesis of secondary metabolites, oxidative phosphorylation, and biosynthesis of cofactors were notably enriched. Genetic information processing was also significantly represented (12 pathways). Environmental information processing pathways, including ABC transporters and two-component systems, accounted for 40 protein annotations, underscoring mechanisms for environmental sensing and molecular transport. Cellular processes such as the cell cycle, mitophagy, and quorum sensing involved 27 proteins across 8 pathways. Despite R. typhi being a bacterial pathogen, 60 proteins were associated with 37 human disease pathways due to conserved or homologous proteins, including those involved in cancer, neurodegeneration, and infectious disease pathways. Overall, the essential proteome of R. typhi is heavily skewed toward metabolic and information processing functions, reflecting the critical systems needed for its intracellular parasitic lifestyle. To prevent negative effects, medications should be selectively lethal against R. typhi rather than against human and beneficial microbes in the gut. Hence, human homologs were subtracted from the dataset. Out of the 426 essential proteins pool, 54 were non-homologous to the gut microbial proteome and 153 were found to be non-homologous to the human proteome (Supplementary Fig. 4). Among this protein set, 55 were associated with general metabolic pathways, while 13 were involved in the biosynthesis of secondary metabolites (Supplementary Table 2). These numbers are significantly reduced compared to the broader essential proteome (Supplementary Table 1), reflecting the subtraction of proteins with human homologs. Notably, pathways like MAPK, HIF-1, AMPK, and PI3K-Akt, glucagon and estrogen signaling etc. were absent. This reflects a refined subset focused primarily on core metabolic processes, fundamental genetic machinery, pathogen-specific systems and minimal involvement in host-like pathways. This subset likely represents high-priority therapeutic targets due to the lower risk of off-target effects in the human host. In addition, multiple proteins were associated with pathogen-specific functions such as lipopolysaccharide biosynthesis, peptidoglycan biosynthesis, flagellar assembly, bacterial secretion systems, ABC transporters, two-component signal transduction systems, antimicrobial resistance pathways, including vancomycin and β-lactam resistance. These pathways are critical for R. typhi virulence, cell wall integrity, motility, and adaptation to host-imposed stress conditions, making the associated proteins high-value candidates for novel drug development. This set of proteins was subjected to druggability investigation and 17 targets were found as the most druggable. IPP isomerase also called isopentenyl-diphosphate delta-isomerase coded by the fni gene was selected as a drug target. It is an important cytoplasmic protein that possesses isomerase catalytic activity having FMN, NADPH, and Mg+ 2 as cofactors [73]. It is involved in the biosynthesis of isoprenoids by catalyzing the 1,3-allylic rearrangement of the homoallylic substrate isopentenyl (IPP) to its allylic isomer, dimethylallyl diphosphate (DMAPP) [74, 75]. IPP isomerase plays a vital role in the terpenoid backbone biosynthesis pathway (KEGG pathway ID: rty00900) and is integral to other key metabolic pathways (KEGG pathway IDs: rty01100, rty01110). The COG (COG1304, COG1443) and GO (GO:0004452) annotations further substantiate its involvement in isopentenyl-diphosphate isomerase activity, underscoring its essential role in cellular metabolic functions. Notably, IPP isomerase is a critical component of both the mevalonate and non-mevalonate pathways for C5 isoprenoid biosynthesis (KEGG modules M00095 and M00096), which are crucial for the metabolic processes of R. typhi. IPP isomerase catalyzes the conversion of IPP to dimethylallyl pyrophosphate via an antarafacial hydrogen shift, essential in isoprenoid biosynthesis [76]. The reaction proceeds through a protonation-deprotonation mechanism, forming a transient carbocation intermediate, followed by removal of a pro-R proton from C2 to generate the final double bond [77]. Although the identities of the specific amino acid residues acting as the proton donor and acceptor remain to be conclusively determined, the overall mechanism is well-supported and highlights the role of enzyme in stereospecific hydride shifts within the isoprenoid biosynthetic pathway. As the terpenoid backbone biosynthesis pathway is involved in various cellular functions such as membrane integrity, signaling molecules, and energy production. Disruption of this pathway through inhibition of IPP isomerase could severely compromise survival and virulence of R. typhi. The KEGG Orthology (KO: K01823) annotation further confirms that the enzyme is highly conserved, which further emphasizes its importance as a broad-spectrum target. Humans primarily utilize the mevalonate pathway and lack the bacterial form of IPP isomerase for biosynthesis of isoprenoids [78]. This fundamental metabolic difference provides a clear therapeutic window, as inhibition of the bacterial enzyme can selectively impair pathogen viability without interfering with host metabolism. Moreover, IPP isomerase is indispensable for R. typhi survival, acting as a metabolic choke point whose inhibition disrupts essential isoprenoid production, leading to bacterial growth arrest or death. This justifies the explicit selection of IPP isomerase as a promising and specific drug target against R. typhi. Virtual screening The compound library substances complied Lipinski’s guideline of five and also fulfilled Opera’s lead-like criteria [79]. The top hits showing the best binding affinity were ZINC01482946 (affinity value: -7.6), ZINC12494625 (affinity value: -7.4), and ZINC06067114 (affinity value: -7.2) (Supplementary Fig. 5). All three ligands (Fig. 3A, C, E) made interactions with 21 residues each (Fig. 3B, D, F). ZINC12494625 made the largest number of hydrogen bonds (Table 2). Lys194, Asn127, Ser68, and Val196 were identified as recurrent interaction hotspots across multiple ligands, suggesting their central role in stabilizing inhibitor binding. Several hydrogen bonds with distances below 3.2 Å indicate strong and specific contacts, which may impede substrate access to the catalytic pocket. Ionic interaction of Lys194 with ZINC12494625 further suggests enhanced affinity. 2D depiction of (A) ZINC01482946 structure (B) IPP isomerase-ZINC01482946 interaction (C) ZINC12494625 structure (D) IPP isomerase-ZINC12494625 interaction (E) ZINC06067114 structure (F) IPP isomerase-ZINC06067114 interaction Table 2Interaction details of hydrogen and ionic bonds for shortlisted top-scoring Streptomycin metabolitesZINC IDLigandReceptorInteractionDistanceE (kcal/mol)Affinity scoreZINC01482946N12OD1Asn127H-donor3.23-7.6C22OVal196H-donor3.21O26SDMet70H-donor3.45O11NSer99H-acceptor3.27O11CBSer99H-acceptor3.33O14ND2Asn127H-acceptor2.97O14CDLys194H-acceptor3.41O26CBAla295H-acceptor3.08ZINC12494625O18CAGly197H-acceptor3.43-7.4O18CAGly273H-acceptor3.00O19NZLys194H-acceptor4.02O19CBAla219H-acceptor3.45O23OGSer68H-acceptor3.48O23CBSer272H-acceptor3.65O23NGly273H-acceptor3.72O23CBAla295H-acceptor3.83O24OGSer68H-acceptor3.06O24NZLys194H-acceptor3.06O24OGSer272H-acceptor2.72O24NZLys194ionic3.06ZINC06067114O14NZLYS194H-acceptor2.99-7.2O14CAVAL196H-acceptor2.93C215-ringHIS157H-pi3.70 Interaction details of hydrogen and ionic bonds for shortlisted top-scoring Streptomycin metabolites The predictions for the binding affinities of the compounds were validated computationally using Vinardo, DiffDock, and CNN scores. The Vinardo score indicated strong interactions for ZINC12494625, suggesting favorable binding (Supplementary Table 3). ZINC01482946 demonstrated a moderate Vinardo binding score (-1.53 kcal/mol) and relatively high CNN score (0.695), consistent with geometrically plausible poses and chemically compatible interactions within the binding pocket. ZINC06067114 displayed weaker Vinardo scores (ranging from 0.165 to 6.787), in line with its lower CNN-based affinity estimates (4.922 to 5.049). DiffDock assigned ZINC01482946 a lower confidence value (-2.75) despite favorable Vinardo and CNN scores. This is not contradictory as DiffDock confidence reflects the stability and reproducibility of sampled poses produced by its diffusion model, rather than direct binding free energy. For ZINC01482946, the diffusion process generated more variable poses with reduced convergence, lowering confidence even when Vinardo and CNN scoring highlighted a chemically favorable binding mode. Together, these methods provide complementary dimensions of IPP isomerase-ligand binding. DiffDock confidence reflects the robustness of pose prediction and sampling reproducibility [80], whereas the Vinardo energy score provides an empirical estimate of binding affinity [81], and the CNN score captures favorable interaction patterns while also incorporating affinity-like characteristics [82]. This multi-pronged evaluation confirms ZINC01482946 as a promising candidate based on CNN and Vinardo scores, even though its lower DiffDock confidence suggests structural variability in predicted poses. Overall, the combination of classical and ML methods offers a balanced approach to evaluating compound efficacy and stability, underlining the importance of multi-faceted validation in computational drug discovery. Molecular weight for ZINC01482946, ZINC12494625, and ZINC06067114 was 30.42, 379.30, and 369.43 Daltons, respectively. Molecular weight can influence drug absorption, metabolism, and pharmacokinetics, and molecules with weight ≤ 500 Daltons are favored [83]. Molecular polar surface area (PSA) was 110.66 Ų, 141.40 Ų, and 75.66 Ų for ZINC01482946, ZINC12494625 and ZINC06067114 respectively. PSA is a highly practical factor for predicting the characteristics of drug transport and is the total of a molecule’s polar atoms on the surfaces (often nitrogen, oxygen, and connected hydrogens) [84]. High Topological Polar Surface Area (TPSA) values indicate a stronger polar character of the molecule and the potential to interact with water, which is crucial for effective drug delivery. Orally active drugs that are intended to be absorbed efficiently through the gastrointestinal (GI) tract often adhere to the guideline that their TPSA should not exceed approximately 120 Ų [85]. This value is used as a rough estimate to ensure that the drug molecules have a balanced hydrophobic and hydrophilic character, which can contribute to optimal absorption. The top scoring molecule ZINC01482946 did not exceed this value. ADMET profiling of top compounds To determine compound safety and avoiding cross-reaction of prioritized phytochemicals, the essential stage in the drug discovery pipeline is ADMET profiling. The water solubility of the compounds was low (Table 3), so to improve this trait, cyclodextrin complexation was employed using AI based webserver. The results demonstrated that SBE-β-CD (sulfobutylether-β-cyclodextrin) provided the most favorable complexation free energy (ΔG) values across all three compounds (Supplementary Table 4). These findings suggest that SBE-β-CD is the most effective cyclodextrin for enhancing the solubility of these compounds, likely through stronger inclusion complex formation, making it a promising formulation choice for improving the bioavailability of these ligands. Table 3ADMET parameters of the top-scoring compoundsPropertyModel nameUnitZINC01482946ZINC12494625ZINC06067114AbsorptionWater solubilityNumeric (log mol/L)-4.208-3.947-5.082Caco2 permeabilityNumeric (log Papp in 10 –6 cm/s)1.2210.9811.405Intestinal absorption (human)Numeric (% Absorbed)95.29267.08499.891Skin PermeabilityNumeric (log Kp)-3.068-2.35-2.956P-glycoprotein substrateCategorical (Yes/No)NoNoNoP-glycoprotein I inhibitorCategorical (Yes/No)YesYesYesP-glycoprotein II inhibitorCategorical (Yes/No)NoNoYesDistributionVDss (human)Numeric (log L/kg)-0.346-0.4430.081Fraction unbound (human)Numeric (Fu)0.0640.2510.038BBB permeabilityNumeric (log BB)-0.778-1.257-0.348CNS permeabilityNumeric (log PS)-2.748-2.935-2.036MetabolismCYP2D6 substrateCategorical (Yes/No)NoNoNoCYP3A4 substrateCategorical (Yes/No)YesYesYesCYP1A2 inhibitorCategorical (Yes/No)NoNoNoCYP2C19 inhibitorCategorical (Yes/No)YesNoYesCYP2C9 inhibitorCategorical (Yes/No)NoNoNoCYP2D6 inhibitorCategorical (Yes/No)NoNoNoCYP3A4 inhibitorCategorical (Yes/No)YesNoNoExcretionTotal ClearanceNumeric (log ml/min/kg)-0.085-0.2731.051Renal OCT2 substrateCategorical (Yes/No)NoYesNoToxicityAMES toxicityCategorical (Yes/No)NoNoNoMax. tolerated dose (human)Numeric (log mg/kg/day)0.2050.2760.612hERG I inhibitorCategorical (Yes/No)NoNoNohERG II inhibitorCategorical (Yes/No)NoNoNoOral Rat Acute Toxicity (LD50)Numeric (mol/kg)2.3952.8081.688Oral Rat Chronic Toxicity (LOAEL)Numeric (log mg/kg bw/day)1.6560.1242.089Skin SensitizationCategorical (Yes/No)NoNoNoT. Pyriformis toxicityNumeric (log ug/L)0.8550.580.978Minnow toxicityNumeric (log mM)-0.097-1.643-0.004 ADMET parameters of the top-scoring compounds Drug molecules that are more widely distributed in different bodily fluids and have fewer adverse effects on the biological system are advantageous. The gastrointestinal permeability of ZINC06067114 was higher than the other two selected compounds whereas none of the three were p-glycoprotein substrates (Table 3). Similarly, these compounds did not exhibit blood-brain barrier permeability and when tested against the cytochromes CYP2D6, CYP2C19, CYP3A4, CYP2C9, and CYP1A2, only ZINC06067114 exhibited inhibition of all three were substrates for CYP3A4. Lipophilicity was moderate for top-scoring compounds (2.44) while ZINC06067114 had a slightly higher lipophilicity (3.96). This means that the prioritized molecules are likely to be well-absorbed through the gastrointestinal (GI) tract, as they can effectively cross cell membranes. However, excessively high lipophilicity can also lead to poor aqueous solubility and potential issues with ADME [86]. The main pharmacokinetic advantages and liabilities of these three top-scoring compounds have been listed in Supplementary Table 5. Neurotoxicity, immunotoxicity, mutagenicity, carcinogenicity, and teratogenicity are the main components of drug toxicity. The selected lead molecules were checked for their toxicity, and the results showed that all of the compounds tested negative for the AMES test and hepatotoxicity, ruling out any possibility of mutagenicity. The T. Pyriformis toxicity of selected ZINC compounds showed values below 1, depicting humans could tolerate this level of toxicity. All of the lead compounds had negative skin permeability as well as sensitivity values (Table 3). Pharmacokinetics also revealed that ZINC01482946 has the highest overall absorption and fraction of the administered dose reaching the systemic circulation (bloodstream) after avoiding the first-pass metabolism in the liver. ZINC06067114 had overall high bioavailability (Table 4). ZINC06067114 seems to have the highest peak concentration, takes the longest to reach that peak, and has the highest overall exposure based on AUC values. However, ZINC01482946 is the most efficiently absorbed and has the highest bioavailability among the three compounds. These data provide insights into the pharmacokinetic profiles of these compounds, which are essential for understanding how they are absorbed, distributed, metabolized, and eliminated in the body. Table 4Mean values obtained for physiology-based pharmacokinetic modeling assayEndpointDescriptionZINC01482946ZINC12494625ZINC06067114Fa [%]Fraction absorbed77.83561.06975.35FDp [%]Fraction dose passing through the portal vein77.72959.2775.205F [%]Bioavailability73.96656.06874.67Cmax [ug/mL]Peak concentration5.90718.122246.372Tmax [h]Time to peak concentration4.60343.689.8382AUC(0-inf) [ng-h/mL]The area under the concentration-time curve from time 0 to infinity163,70078,870861,000AUC(0-t) [ng-h/mL]The area under the concentration-time curve from time 0 to 10 h48,05055,050250,900 Mean values obtained for physiology-based pharmacokinetic modeling assay MD simulation MD simulation results of the ZINC01482946 and ZINC12494625 with IPP isomerase revealed an average RMSD of less than 1.5 Å (i.e. 1.08 and 1.25 Å respectively). For complex with ZINC06067114, it increased above 1.5Å after 90 ns and was on the average 1.20 Å. Overall ZINC01482946 had the lowest RMSD i.e. average 1.08 Å over 100 ns. Overall, ZINC01482946- IPP isomerase complex demonstrated the greatest stability during MD simulations, as evidenced by consistently lower RMSD values compared to the other complexes over the 100 ns trajectory (Fig. 4A). This stability was further supported by the violin plot, which shows that the RMSD distribution for ZINC01482946 is narrower and centered at a lower value than those for ZINC06067114 and ZINC12494625 (Fig. 4B). (A) RMSD of ligand bound IPP isomerase complex over 100 ns (B) Violin plot of RMSD distribution points overlaid on box plot RMSF quantifies the flexibility or mobility of individual residues in a protein throughout MD simulation, representing how much each residue deviates from its average position during the trajectory. Higher RMSF values indicate greater residue flexibility, often corresponding to loops or surface regions, whereas lower values reflect more rigid or stable regions, typically structured elements such as helices or sheets. Average RMSF was 0.64, 0.70 and 0.80 Å for ZINC01482946, ZINC12494625 and ZINC06067114 respectively. Overall, residues exhibit similar flexibility profiles across complexes but ZINC01482946 complex generally shows slightly lower RMSF values, suggesting enhanced protein stability upon binding (Supplementary Fig. 6). Thus, RMSF analysis complements RMSD and distance metrics by providing residue-level insights into dynamic behavior and ligand-induced stability changes. Analysis of the distance between each ligand and the catalytic site of IPP isomerase (Fig. 5) revealed that ZINC01482946 consistently maintained the shortest distance throughout the simulation (clustering around 1–2 Å), indicative of strong and persistent interaction with the active site. Quantification of mean distances showed that ZINC01482946 had the lowest average, suggesting more stable catalytic engagement. ZINC06067114 displayed an intermediate average, and ZINC12494625 exhibited the highest mean with greater variability, reflecting weaker or less stable proximity. Kernel density distributions further supported these observations, with ZINC01482946 tightly clustered at lower distances and indicating high occupancy within the catalytic region. ZINC12494625 displayed a broad distribution with a higher mean, consistent with looser and more transient contacts, while ZINC06067114 showed a bimodal distribution, suggesting occasional but not sustained catalytic site interactions. Collectively, the consistently lower catalytic site distance of ZINC01482946 reinforces its candidacy as a strong IPP isomerase inhibitor, implying a robust and preferential binding geometry. These findings corroborate RMSD results, highlighting superior stability and active site engagement of ZINC01482946 relative to the other compounds. (A) Distance between ligand and catalytic site residues of IPP isomerase over time. (B) Average ligand-catalytic site distance as a function of time. (C) Distance distribution and corresponding kernel density plots ZINC12494625 exhibited the most favorable contributions from both electrostatic and van der Waal’s (VDW) interactions. ZINC01482946, despite variable electrostatics, maintained strong binding through consistent VDW and other stabilizing interactions, while ZINC06067114 formed less stable complexes overall. Thus, ZINC01482946 remains a promising candidate, with binding stability supported by multiple energy components beyond electrostatics. The PCA analysis revealed ligand-specific effects on the conformational dynamics of IPP isomerase. For ZINC01482946, the system formed a broadly distributed cluster along both PC1 and PC2 axes, indicating wide sampling of conformational states with a relatively uniform population (Fig. 6). This suggests stable equilibrium dynamics or transitions among similar conformations. In contrast, ZINC06067114 induced the largest conformational changes, with distinct clusters spread along PC1 and occupying elongated, less compact regions, indicative of pronounced conformational shifts or multiple dominant states. ZINC12494625 displayed two distinct clusters and moderate movement along PC1, revealing transitions between favored conformations. Overall, ZINC01482946 and ZINC12494625 maintained the complex within constrained conformational regions, consistent with more stable or less disruptive binding, whereas ZINC06067114 promoted greater flexibility or potential destabilization. PCA plot for ligand bound IPP isomerase complex, simulated over 100 ns Smaller Rg values generally indicate a more compact and well-folded structure, while larger values suggest greater structural flexibility or unfolding (Supplementary Fig. 7). Average values were 19.39, 19.38 and 19.42 Å for ZINC01482946, ZINC12494625 and ZINC06067114, respectively. ZINC12494625 and ZINC06067114 seemed to have a decreased Rg compared to initial state, meaning that the system was attaining a stable conformation. ZINC01482946 formed occasional ligand-specific hydrogen bonds (0–2) throughout the simulation, indicating transient but potentially stabilizing interactions (Supplementary Fig. 8). In contrast, ZINC06067114 and ZINC12494625 did not form any persistent ligand hydrogen bonds, suggesting weaker direct interactions with the protein. Total protein hydrogen bonds remained relatively stable for all complexes, with ZINC01482946 ranging from 308 to 330, ZINC06067114 from 287 to 323, and ZINC12494625 from 305 to 335. The total SASA for ZINC01482946 fluctuated moderately between ~ 257–265 Ų, reflecting minor conformational adjustments while maintaining overall protein exposure (Supplementary Fig. 9). ZINC06067114 showed a similar SASA range (~ 259–265 Ų) but with slightly greater variability, indicating more flexible protein dynamics. ZINC12494625 displays slightly lower SASA values (~ 251–257 Ų), suggesting tighter packing and reduced solvent exposure. Taken together, ZINC01482946 exhibited the most favorable profile, with occasional ligand hydrogen bonds supporting stable engagement and moderate SASA fluctuations indicating structural stability. ZINC06067114, lacking ligand hydrogen bonds and showing higher SASA variability, may induce greater conformational flexibility and less stable binding. ZINC12494625 maintains a compact protein structure but lacks ligand-specific hydrogen bonds, suggesting constrained protein dynamics but potentially weaker direct ligand interaction. These observations align with RMSD, RMSF, and PCA analyses, supporting ZINC01482946 as the strongest candidate for stable IPP isomerase inhibition. For ZINC01482946 (Supplementary Fig. 10), the total binding energy remained relatively stable over time, with interaction energies ranging between − 50 and − 80 kcal/mol, suggesting stable binding. However, electrostatic energy exhibited moderate fluctuations. ZINC06067114 displayed slightly less negative total energies than ZINC01482946, but with a more consistent interaction profile despite electrostatic oscillations. ZINC12494625 exhibited substantially more negative energies, indicating the strongest binding affinity among three ligands. Across all ligands, van der Waals contributions remained relatively stable with minimal variability, reflecting consistent non-polar interactions. Fluctuations in electrostatic energy suggest dynamic polar interactions, likely arising from ligand-protein rearrangements or flexible binding modes during the simulation. MM/GBSA values were also calculated and the trends in the energy components and delta G values reveal variations in the contributions of van der Waals forces, electrostatic interactions, solvation energies, and gas-phase energies across the complexes (Supplementary Table 6). The ZINC12494625-IPP isomerase complex showed the lowest solvation free energy (G_solv), supporting stability in solvent. In contrast, the ZINC01482946-IPP isomerase complex yielded the most negative ΔG value, suggesting the most favorable binding affinity overall. Discussion R. typhi is a zoonotic pathogen and is responsible for endemic typhus [87]. It has circular chromosome and its complete proteome sequence is publicly available. Till now, no plasmid has been reported in this microbe [88, 89]. The bacterium is typically transmitted to humans through fleas that infest rats and is characterized by flu-like symptoms [10]. Infection by Rickettsia spp. have been linked with the autoimmune conditions ‘multiple sclerosis’ and ‘Guillian- Barre syndrome’. Rocky mountain fever by R. rickettsii has been allied with vasculitis (autoimmune disorder) [90]. Rickettsia spp. presence within endothelial cells leads to their destruction, causing the release of self-antigens. This triggers an immune response characterized by the production of antiphospholipid antibodies and local inflammation, which may contribute to thromboangiitis obliterans, an autoimmune condition marked by inflammation and thrombosis in small and medium-sized arteries [91]. R. prowzakii homolog to human thyrotropin receptor, binding HLA-DR3 allele has been found to trigger autoimmune thyroid [31, 92]. Rickettsia rickettsii presence within endothelial cells leads to the destruction of these cells, resulting in the release of self-antigens. This triggers an immune response characterized by the production of antiphospholipid antibodies and local inflammation, contributing to thromboangiitis obliterans, a condition marked by inflammation and thrombosis in small and medium-sized arteries. Based on this information, molecular mimicry or autoimmunity was probed for R. typhi by mining human homologs and then peptides >= 10 amino acid in length aligned with these homologs. Sequence similarity between these peptides and host proteins suggests potential for immune cross-reactivity, possibly triggering autoimmune responses. In total, 25 proteins (homology >= 50%) exhibited molecular mimics with potential for autoimmune diseases like psoriasis, rheumatoid arthritis etc. This diversity suggests that R. typhi proteins may target a wide range of cellular processes and autoimmune disorders. However, we acknowledge that sequence similarity alone is an imperfect predictor of autoimmune risk, as immune recognition is strongly influenced by the 3D structure of epitopes, their binding specificity to Major Histocompatibility Complex (MHC) molecules, and the biological context of antigen presentation [93, 94]. Linear sequence matches do not necessarily translate into the structural homology required for antibody or T cell receptor recognition. Moreover, in silico predictions cannot confirm in vivo processing, presentation, or immunogenicity of shared motifs, many of which may be false positives. Additionally, autoimmune diseases arise from complex interactions beyond mimicry, including genetic and environmental factors and failures in immune tolerance [95]. Hence, sequence similarity can provide a valuable hypothesis-generating framework but its implications for autoimmunity must be validated through structural analyses, epitope mapping, and experimental immunological assays to establish true relevance to disease pathogenesis. Assays like Enzyme-Linked Immunosorbent Assay (ELISA), Western blotting, and flow cytometry can be employed to detect autoantibodies or autoreactive T cells against the putative mimicking epitopes in patient samples or immunized models [96]. Functional assays such as cytokine release, T cell proliferation assays, and in vivo animal models may also validate the biological consequence of mimicry, establishing causal links to autoimmune pathology [95, 97]. Longitudinal studies monitoring mimicry-driven immune responses can also clarify whether mimicry is a trigger or an epiphenomenon [98, 99]. Another aim of this study was to identify drug targets in R. typhi that are selectively lethal to the bacteria while sparing human cells and beneficial gut microbes. We took the complete proteomes of the species (n = 4) and mined drug targets using subtractive proteomics approach. Among them, IPP isomerase, encoded by the fni gene, was considered a promising drug target for downstream analysis. This cytoplasmic protein is involved in the biosynthesis of isoprenoids [74, 75]. It was conserved more than 80% in 22 Rickettsia spp. (Supplementary Fig. 11) and was most similar to R. prowazekii (Supplementary Fig. 12). In addition, IPP isomerase inhibition is therapeutically promising due to its essential role in Rickettsia spp. metabolism, unique evolutionary retention amidst genome reduction, and its potential for selective targeting. Additionally, host-directed therapies like statins inhibit upstream host isoprenoid biosynthesis and have demonstrated inhibitory effects on Rickettsia growth. This further highlighting the therapeutic potential of manipulating isoprenoid pathway. Current anti-rickettsial therapies, such as tetracyclines, function by disrupting protein synthesis via inhibition of the 30 S ribosomal subunit [100]. While effective, these antibiotics can affect host mitochondrial ribosomes due to their prokaryotic ancestry, contributing to potential host toxicity [101]. Similarly, cell wall biosynthesis inhibitors are less effective against Rickettsia, which have a reduced and atypical peptidoglycan structure, limiting the utility of β-lactam antibiotics. Consequently, IPP isomerase inhibition offers an attractive alternative, especially in scenarios where resistance, allergies, or contraindications limit the use of first-line treatments. The therapeutic limitations of currently used antibiotics alongside the growing threat of antimicrobial resistance, highlight the urgent need for novel treatments. Screening natural products and exploring new chemical space for treating R. typhi infections could lead to the discovery of drugs that are less prone to resistance and a more favorable safety profile with fewer adverse effects on the host, compared to some traditional antibiotics. Streptomyces metabolites are a rich source of compounds with diverse applications in medicine, agriculture, biotechnology, and ecological research [102, 103]. Researchers have studied these metabolites for their potential as pharmaceuticals, as they often have complex chemical structures and unique biological activities [104]. Streptomyces metabolites serve as a valuable resource for exploring new therapeutic agents and are reported to have anticancer, antifungal, or immunosuppressive properties, making them valuable in drug discovery and development [105]. Some examples of antibiotics derived from Streptomyces spp. include pluramycin [106], hanneumycin [107], ravidomycin [108], nonactin [109] etc. These antibiotics have exhibited antibacterial activity. To explore novel antibacterial candidates against R. typhi, a structurally diverse subset of Streptomyces spp. metabolites was selected for virtual screening against IPP isomerase of R. typhi. IPP isomerase represents a promising target for the development of novel antimicrobial agents due to its essentiality in microbial metabolism and its absence in the human host. Receptor-based screening using AutoDock Vina revealed ZINC01482946, ZINC12494625, and ZINC06067114 to have good binding affinities against this enzyme. The findings were validated using ML methods because integrating classical docking with ML-based methods leverages the complementary strengths of both approaches. AutoDock Vina rapidly generates plausible ligand conformations using an empirical scoring function, while ML models refine these predictions by learning complex interaction patterns from large datasets. This hybrid strategy offers a balanced framework that combines speed, interpretability, and accuracy, ultimately improving the robustness of structure-based drug discovery pipelines. The observed stability of the ZINC01482946 complex through MD simulation strongly supports its potential as a drug candidate, as persistent low RMSD values suggest a robust and persistent binding orientation, which is desirable for inhibitory efficacy against the target. ZINC06067114 showed high bioavailability and peak concentration, while ZINC01482946 exhibited efficient absorption and high bioavailability. It also exhibited TPSA < 120 Ų, which generally corresponds to a more hydrophobic and less polar molecule. Such molecules are more likely to cross lipid-rich biological membranes, including the blood brain barrier (BBB), and are better absorbed through the GI tract. Pharmacokinetic profiling offers critical information on how the drug molecules are processed in the human body and is invaluable for optimizing drug formulations and dosing regimens. However, it is important to note that these are general guidelines, and other factors, such as molecular weight, charge, and specific interactions with transporters or receptors, also play critical roles in determining pharmacokinetic and pharmacodynamics of a compound. Effective delivery to intracellular pathogens such as R. typhi requires compounds to not only cross host cell membranes but also remain stable and bioactive in the intracellular environment. These factors complicate both dosing strategies and clinical outcomes. Overcoming these limitations often requires formulation enhancements, such as the use of solubility-enhancing agents, nanoparticle systems, or inclusion complexation strategies like cyclodextrins, which can improve drug solubility, stability, and targeted delivery. Owing to hydrophobic inner cavities and hydrophilic outer surfaces, cyclodextrins are well-suited to form inclusion complexes with compounds of low or moderate solubility, thereby increasing their apparent solubility and dissolution rate. SBE-β-CD demonstrated promising complexation potential and warrants further investigation. The strengths of this study include leveraging ML validation of classic docking, to enhance prediction accuracy of inhibitor screening against R. typhi. Additionally, the PBPK modeling provided preliminary insights into pharmacokinetics, strengthening the translational relevance. Furthermore, formulation strategies, such as coupling candidate compounds with cyclodextrins, highlight potential improvements in bioavailability. The mimicry analysis represents a novel application of sequence alignment to explore R. typhi and human protein homologs, aiming to uncover potential autoimmune triggers or immune evasion strategies. Although not directly targeted in this study, mimicry-associated proteins offer valuable mechanistic insights and represent promising therapeutic targets for future investigation. The computational findings in this study represent an important early step in drug discovery against R. typhi, allowing the prioritization of candidates with a higher likelihood of modulating target protein function. This is an essential aspect of combating bacterial pathogenesis and the identification of potential inhibitors, such as ZINC01482946, provides a promising starting point for translational research. Clinically, if proven effective, ZINC01482946 could offer a novel adjunct or alternative to current R. typhi treatments, particularly valuable in cases of antibiotic resistance or treatment failure. Immunocompromised individuals, patients with allergies to first-line antibiotics, or those in endemic regions with limited healthcare access might benefit first from such a targeted therapeutic. Despite these encouraging prospects, it is important to recognize that these insights are based on in silico analyses, which are informative but inherently predictive and subject to limitations. The limits of this study include potential false positives in target identification as well as ligand binding, overestimation of binding affinities, and the inability to fully capture complex cellular or organismal dynamics. In molecular docking, false positives often arise from inaccurate scoring functions that fail to capture the true energetics of protein-ligand interactions, including solvation effects, entropic contributions, and protein flexibility. The assumption of rigid or semi-flexible proteins further compounds this issue, as it overlooks conformational changes critical for accurate binding predictions. Inadequate modeling of desolvation penalties, especially for polar groups lacking proper hydrogen bonding may lead to ligands being incorrectly ranked as strong binders. The lack of standardized validation datasets and limited sampling of binding modes or protein conformations further increase inconsistency and false positives. In large-scale virtual screens, small predictive errors can propagate, yielding many top-ranked compounds that prove inactive experimentally. While ADMET and PBPK modeling provide preliminary insights into pharmacokinetics, metabolism, and potential toxicity, these predictions cannot replace experimental validation as well. These computational limitations highlight the need for careful experimental validation and integration with biological context to ensure translational relevance. Therefore, rigorous in vitro and in vivo studies are required to confirm the biological relevance and therapeutic potential of the predicted inhibitors. To advance the therapeutic potential of ZINC01482946 as an IPP isomerase inhibitor, future research should prioritize comprehensive experimental validation using both cellular and animal models. In parallel, exploring combination therapies is essential. Evaluating the synergistic interaction between ZINC01482946 and established antibiotics, such as doxycycline or azithromycin, could potentiate antimicrobial activity, reduce effective dosages, and minimize host toxicity. This combinatorial approach may also offer a viable strategy to curb the emergence of drug resistance. Given the iterative nature of drug discovery, if the compound fails to meet efficacy or safety, scaffold modification and re-testing could be carried out, ensuring a robust pipeline for candidate optimization. Another limitation of this study is that computational approaches for identifying molecular mimicry can generate false positives. Such analyses often yield numerous candidate epitopes, many of which may lack biological relevance, complicating experimental validation and potentially obscuring the identification of truly pathogenic mechanisms. Many in silico methods focus on sequence-based mimicry detection, which may miss conformational or structural mimicry. Predicting mimicry at the conformational epitope level is particularly challenging. Only mimicry in surface-exposed protein regions is likely to have functional effects, such as triggering autoimmune responses. Computational predictions cannot definitively determine immune consequences, as factors like T cell regulation and prior immune education play a role. The absence of standardized criteria for validating mimicry predictions further contributes to inconsistencies across studies because the effectiveness of mimicry prediction depends on the availability of comprehensive genomic, proteomic, and structural data for both hosts and pathogens. A lack of high-quality proteomes and resolved protein structures, particularly for less-studied species like R. typhi, limits accurate detection of structural similarities. Nonetheless, the mimicry analysis provides valuable complementary insights into R. typhi pathogenesis, highlighting proteins that modulate host immune responses and may serve as future therapeutic targets. Although the mimicry-associated proteins identified were not directly targeted in the current drug screening pipeline, focusing on these regions could reveal vulnerabilities that enable selective inhibition of bacterial proteins while minimizing off-target effects on human homologs, thereby linking autoimmune mimicry insights to rational drug design. Conclusion Pan-proteome analysis revealed that R. typhi possesses an open pan-proteome, reflecting its evolutionary adaptability and horizontal gene transfer. The investigation of molecular mimicry within the R. typhi proteome shed light on host-pathogen interactions and provided insights into potential autoimmune triggers, offering further opportunities for diagnostic marker development. The conserved proteome was made of 803 proteins and functional annotation indicated that many of these proteins were enriched in essential processes such as translation, replication, repair, and energy metabolism, underscoring their critical role in bacterial survival. The subtractive screening of yielded a pool of 426 essential, non-redundant candidates. Among these, 17 proteins had a strong druggability potential and 19 were associated with virulence properties. IPP isomerase was taken as an essential drug target and a virtual screening workflow was used to identify inhibitors from Streptomyces spp. metabolite library. The compound ZINC01482946 was identified as the best binding inhibitor, with a favorable ADMET profile. Together, these findings present a multi-layered strategy for therapeutic exploration of R. typhi, integrating pan-proteomics, subtractive proteomics, and rational drug design. To the best of our knowledge, this is the first report of Streptomyces-derived inhibitors contributing to early-stage anti-rickettsial drug discovery. The lead inhibitor warrants further validation through in vitro and in vivo studies, offering a foundation for the development of novel anti-rickettsial therapies and improved management of typhus infections. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors extend their appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through Large Research Project under grant number RGP2/95/46. Abbreviations Absorption, Distribution, Metabolism, Skin Penetration, Excretion Bacterial and Viral Bioinformatics Resource Center Bacterial Pan Genome Analysis Tool Basic Local Alignment Search Tool Blood Brain Barrier Cluster of Essential Genes Clusters of Orthologous Groups Convolutional Neural Network Database of Essential Genes Dimethylallyl Diphosphate Enzyme-Linked Immunosorbent Assay Generalized Born Gastrointestinal Groningen Machine for Chemical Simulation Isopentenyl Isopentenyl Pyrophosphate Kyoto Encyclopedia of Genes and Genomes Linear Constraint Solver Machine Learning Major Histocompatibility Complex Molecular Dynamics Molecular Mechanics / Generalized Born Surface Area Multiple Sequence Comparison by Log- Expectation Optimized Potentials for Liquid Simulations - All Atoms Particle Mesh Ewald Physiological Based Pharmacokinetic Principal Component Analysis Radius of Gyration Ribonucleic Acid Root Mean Square Deviation Root Mean Square Fluctuation Solvation Free Energy Soluble Liver Antigen/Liver-Pancreas Solvent-Accessible Surface Area Topological Polar Surface Area Unweighted Pair Group Method with Arithmetic Mean Universal Protein Resource Van Der Waal’s Author contributions Z.B., F.A.M.A., I.A., S.K. conceived and designed the study. Z.B., S.S.H, M.I., C.R.W. performed the experiments. Z.B., S.S.H., M.S., Y.W., wrote the first draft of manuscript. C.R.W., Y.W., M.I., H.A.O. and F.A.M.A. validated findings. M.I., I.A., Y.W., F.A.M.A., H.A.O. visualized, curated data and edited the draft. I.A., F.A.M.A., H.A.O., S.K., supervised the study, obtained funding and provided resources/software. Funding The authors acknowledge the National Research Foundation of Korea (Grant number: RS-2022-NR066818). Dr. Ibrar Ahmed is a recipient of the Brain Pool Fellowship by the National Research Foundation of Korea (Grant Number RS-2023-00223245). This work was funded by the Deanship of Research and Graduate Studies at King Khalid University through Large Research Project under grant number RGP2/95/46. Data availability The dataset supporting the conclusions of this article are included within the article and its additional files. Declarations Ethical consent Not applicable. Patient consent Not applicable. Competing interests ZB and IA are employed by Alpha Genomics Private Limited. CRW is employed by Shing Huei Group. SSH, YW, HAO, FAMA, MS, MI, SK declare that they have no conflict of interest. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Zarrin Basharat, Email: zarrin.iiui@gmail.com. Seil Kim, Email: stapler@kriss.re.kr. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The dataset supporting the conclusions of this article are included within the article and its additional files.