Clinical peptidomics for respiratory diseases: matrices, workflows, and translation towards treatable traits, with a focus on COPD Authors: Qingyu Zhou, Yahui Shen Published: 2026 ABSTRACT AbstractBackgroundPeptidomics is an emerging tool for biomarker discovery. By capturing end products and active peptides generated during protein breakdown, it helps reveal short peptides linked to disease. This narrative review centers on chronic obstructive pulmonary disease (COPD) and synthesizes recent advances in respiratory peptidomics across patient-accessible matrices and laboratory workflows toward treatable-trait translation.Main bodyWe conducted a structured literature search of PubMed, Embase, and Web of Science (January 2000–September 2025, with emphasis on the most recent five years), prioritizing mass-spectrometry–based discovery and targeted verification in human biospecimens. We distinguish clinical endogenous peptidomics, immunopeptidomics, and degradomics as complementary approaches in respiratory disease. Six representative peptide classes were compared across pre-analytical handling, enrichment strategies, and MS identification, building an evidence map for COPD, asthma, lung cancer, and pulmonary fibrosis. Using this map, we discuss matrix–technology fit and recurrent biological signals—airway inflammation, extracellular matrix turnover, and host–pathogen interaction—that show promise for disease subtyping and early diagnosis. For translation, we outline a stepwise pathway: (i) harmonized sampling and internal-standard–driven quality control; (ii) transparent modeling with calibration and decision-curve analysis; and (iii) multicenter external validation. We further consider integration with proteomics and breathomics, emerging peptide-drug leads, and open sharing of data and code to improve reproducibility and transferability.ConclusionsPeptidomics is poised to contribute clinically actionable biomarker panels in respiratory disease, with near-term opportunities in COPD phenotyping and exacerbation risk assessment using sputum and blood. Broad adoption will depend on standardized pre-analytics, feasible targeted assays in routine laboratories, robust external validation, and transparent model calibration.Graphical abstract Keywords: Peptidomics, COPD, Respiratory diseases, Biomarkers, Mass spectrometry, Treatable traits, Peptide therapeutics Background According to the Global Burden of Disease (GBD) 2021 study, chronic respiratory diseases cause approximately 4.41 million deaths annually and 26.2 million years lived with disability (YLDs), posing a long-standing burden on global health [1]. COPD is among the largest contributors to this burden. In 2020, its global prevalence was approximately 10.6%; although the proportion may decline slightly, the absolute number of cases is projected to reach about 592 million by 2050 [2, 3]. Early COPD is often clinically silent. Diagnosis still relies almost entirely on spirometry, which provides limited insight into underlying molecular heterogeneity, and robust biofluid biomarkers for phenotyping and risk stratification are still lacking. To address this issue, COPD biomarker research has utilized multi-omics studies to map this heterogeneity. Genomic and transcriptomic analyses have identified genetic signatures linked to faster lung function decline, inflammatory responses, and exacerbation risk [4, 5]. Proteomic profiling of lung tissue, bronchoalveolar lavage fluid (BALF), sputum and serum has highlighted inflammatory changes and shifts in extracellular matrix (ECM) proteins [6, 7], while metabolomics has revealed systemic metabolic changes related to oxidative stress, cachexia, and cardiometabolic conditions [8]. However, most of these measurements are static and cannot capture short-lived processes such as ECM remodeling, which are critical in COPD progression. In addition, patient heterogeneity, technical complexity and costs, difficulties in data integration, and the lack of standardization and clinical validation still limit the widespread use of multi-omics in COPD research. Peptidomics, by bridging genomics, proteomics and phenotypic biology, captures the dynamic repertoire of endogenous bioactive peptides and offers a molecularly informed avenue for disease assessment [9]. Over the past two decades, peptide detection has progressed from matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) and surface-enhanced laser desorption/ionization time-of-flight (SELDI-TOF) fingerprinting to identification by liquid chromatography tandem mass spectrometry (LC-MS/MS) and to quantification at high throughput using data-independent acquisition (DIA), exemplified by sequential window acquisition of all theoretical fragment ion spectra (SWATH). These advances have improved reproducibility across batches and enabled larger cohort studies [10, 11]. The concept of treatable traits, defined as measurable, clinically relevant and modifiable patient characteristics that can be targeted with specific interventions, has been introduced to move COPD management beyond spirometry-based categories toward a more personalized, mechanism-driven approach [12–14]. In this framework, airway disease is deconstructed into pulmonary, extrapulmonary and behavioural traits that are independent of diagnostic labels. In COPD and asthma, proposed traits include airflow limitation, type-2 or neutrophilic airway inflammation, frequent exacerbations, ECM remodelling, and systemic factors such as cardiovascular and metabolic comorbidities [15]. Within this paradigm, there is a clear need for biomarkers that specifically report on protease activity, matrix turnover and inflammation related to these traits and that can be monitored longitudinally in accessible biofluids. In this narrative review, we focus on respiratory diseases, with an emphasis on COPD; summarize LC-MS/MS-based peptidomic workflows from sampling and acquisition to bioinformatic analysis and targeted validation; synthesize peptidomic evidence from the last five years across key matrices, including serum, sputum and BALF, relate differential peptide signatures to COPD-related treatable traits, and finally outline practical steps needed to accelerate the clinical translation of peptide biomarkers. Literature search and study selection We conducted a narrative review based on a structured literature search. We searched PubMed, Embase and Web of Science for articles published between January 2000 and September 2025, with a particular focus on the most recent 5 years. The search used terms for peptidomics, clinical peptidomics, immunopeptidomics, and degradomics. These terms were combined with terms for respiratory diseases, such as chronic obstructive pulmonary disease, asthma, bronchiectasis, pneumonia and acute respiratory distress syndrome. Searches were limited to human studies, English language publications and journal articles. Titles and abstracts were screened to retain studies in respiratory diseases that applied peptidomics in clinical samples. To identify clinically oriented work, we also repeated the search with additional terms such as biomarker, phenotype and treatable trait, and we checked the reference lists of key original and review papers to find additional studies. We included original human studies that used mass spectrometry based peptidomics or peptide profiling workflows in clinical samples from patients with respiratory diseases. Accepted matrices were blood, serum, plasma, sputum, bronchoalveolar lavage fluid, exhaled breath condensate and lung tissue. In the full text, we confirmed that the peptidomics workflows were based on LC-MS or LC-MS/MS, or other mass spectrometry platforms, and that the studies reported peptide patterns, candidate biomarkers or relations to clinical phenotypes or potential treatable traits such as disease severity, exacerbations, symptoms or comorbidities. For the formal study selection, we excluded animal studies, in vitro work, articles without peptide level mass spectrometry data, and non original articles such as reviews, editorials and conference abstracts. Review articles, editorials, and conference abstracts were used for background but were not counted in the final number of included studies. The structured search for clinical peptidomics studies first identified 234 records; after removal of 61 duplicates and screening of titles and abstracts, 173 records were screened by title and abstract. 99 full-text articles were assessed for eligibility, and 17 original clinical peptidomics studies in respiratory diseases were finally included in the formal study set for evidence mapping. Methodological or workflow references (e.g., those summarized in Table 2) were cited narratively and were not counted in the included-study total. The selection process is summarized in a PRISMA-style flow diagram (Fig. 1). Table 2COPD-specific peptide signatures with clinical relevance and validation statusPeptide or panelSpecimen & contrastDirectionClinical relevance/treatable traitValidation statusPyroGlu des-Arg fibrinopeptide BSerum; COPD smokers vs. healthy smokers↑ in COPD smokers [105]Early COPD diagnostic marker; reflects inflammation/thrombosisDiscovery + internal validation (small test set)HRG-derived peptide(s)Serum; CAL* vs. healthy (also elevated in COPD)↑ in early CAL and in COPD [17]Diagnostic for early airflow limitationDiscovery (LC-MS/MS panel); no external replicationα1-Acid glycoprotein (AGP1) peptideSerum; CAL vs. healthy (also in COPD)↑ in early CAL and in COPD [17]Systemic inflammation marker; discriminates COPDDiscovery (LC-MS/MS panel); no external replicationα1-Antitrypsin (A1AT) peptideSerum; CAL vs. healthy (also in COPD)↑ in early CAL and in COPD [17]Associated with antiprotease response; diagnostic aidDiscovery (LC-MS/MS panel); no external replicationSLPI (secretory leukoprotease inhibitor)Sputum; smokers with COPD vs. without↑ in COPD (higher with worse FEV₁) [92]Correlates with severe airflow limitation (low FEV₁)Discovery (cross-sectional); no external replicationβ-Defensin-1Sputum; smokers with COPD vs. without↑ in COPD (higher with worse FEV₁) [92]Correlates with severe airflow limitation; neutrophilic inflammationDiscovery (cross-sectional); no external replicationNeutrophil α-defensins (HNP1–3)Sputum; stable vs. exacerbation↑ during AECOPD (highest in exacerbation [106]Predicts COPD exacerbation risk; indicates disease severityDiscovery (single-center); no external replicationβ-Defensin-2 (hBD-2)Sputum; COPD patients (exacerbators vs. stable)↓ in patients with future exacerbations [107]Low levels indicate impaired innate defense; linked to exacerbationsDiscovery (prospective study); no external replicationC1M & C6M (collagen fragments)Serum; COPD patients (rapid vs. slow decliners)↓ in rapid FEV₁ decliners [108]Reflects accelerated ECM turnover (emphysema progression)Discovery + internal validation (ECLIPSE cohort)AECOPD – acute exacerbation of COPD; *CAL – chronic airflow limitation (pre-COPD at-riskstage).References [103–106] are newly added (not cited in the manuscript) COPD-specific peptide signatures with clinical relevance and validation status AECOPD – acute exacerbation of COPD; *CAL – chronic airflow limitation (pre-COPD at-riskstage).References [103–106] are newly added (not cited in the manuscript) PRISMA flow diagram of the literature search and study selection for clinical LC-MS–based peptidomics studies in respiratory diseases Conceptual landscape of clinical peptidomics Peptidomics in respiratory disease can be broadly grouped into three fields: clinical endogenous peptidomics, immunopeptidomics and degradomics. Together, these approaches capture different aspects of peptide biology and offer complementary views on immune responses, inflammation and tissue remodelling in conditions such as COPD, lung cancer and viral pneumonia. Clinical endogenous peptidomics Clinical endogenous peptidomics focuses on the analysis of low molecular weight peptides in biofluids such as plasma, serum, BALF, and sputum, derived from the degradation of ECM components, antimicrobial peptides, and other biologically active molecules [16]. These peptides indicate of ongoing inflammation, tissue injury, and ECM remodeling, which are key features in diseases like COPD. In plasma, peptide panels derived from α1-antitrypsin and histidine-rich glycoprotein can distinguish chronic airflow limitation and COPD from healthy controls, and these signatures correlate with lung function decline and exacerbation risk [17]. For these reasons, this review mainly focuses on this clinical endogenous peptidomics axis in respiratory biofluids, which is currently the most directly translatable to routine patient stratification and treatable-trait assessment. Immunopeptidomics Immunopeptidomics profiles the peptides presented by major histocompatibility complex (MHC, HLA) molecules on the surface of structural and immune cells, thereby defining the antigen repertoire seen by T cells [18]. In oncology and infectious diseases, it is widely used to monitor antigen presentation and guide vaccine or immunotherapy development by mapping tumor- or pathogen-derived epitopes [19]. DIA based LC–MS workflows and curated spectral libraries, such as those collected in the SysteMHC Atlas, are already well established in this field [18]. Compared with clinical endogenous peptidomics, which measures soluble degradation peptides in biofluids, immunopeptidomics mainly uses tissues or cells and focuses on antigen presentation to T cells, rather than circulating peptide fragments. Degradomics Degradomics focuses on protease–substrate networks reconstructed from peptide data. In clinical peptidomics, this means taking panels of endogenous peptides and inferring which proteases most likely generated them. Tools such as Proteasix, built on curated cleavage-site information from resources like MEROPS, map peptide termini back to candidate proteases [20, 21]. In COPD and other chronic respiratory diseases, this approach can be applied to peptides derived from elastin, collagen, mucins and antiproteases to outline the activity of neutrophil elastase, matrix metalloproteinases (MMPs) and related enzymes that drive extracellular matrix destruction and mucus obstruction [22, 23]. When degradomic readouts are interpreted together with fluid-based endogenous peptidomic signatures and transcriptomic or proteomic measurements of the same pathways, peptide fingerprints can be translated into mechanistic insight and potentially treatable protease–antiprotease imbalance. Multi-omics integration, including tools like Data Integration Analysis for Biomarker discovery using Latent components (DIABLO) and Multi-Omics Factor Analysis (MOFA), allows peptidomic data to be analyzed within a unified framework [24, 25]. These tools have demonstrated success in cardiovascular and respiratory disease settings and can be transferred to peptide-centric frameworks. The integration of degradomics data with proteomic and transcriptomic measurements helps reveal disease-specific pathways and identify biomarkers. By linking peptide data with clinical proteomic and transcriptomic data, degradomics provides actionable insights into disease mechanisms and therapeutic targets. Clinical peptidomics workflows Pre-analytical processing, enrichment and key quality metrics Pre-analytical processing and enrichment Pre-analytical processing governs downstream enrichment success and peptide integrity. Across peptide and matrix types, three general principles apply. First, preserve peptide stability: within 30 min of collection, perform cold, rapid centrifugation, adding protease inhibitors and adjusting pH to block secondary proteolysis [26, 27]. Second, mitigate matrix-specific interferences. In plasma or serum, masking of albumin (Alb) or immunoglobulin G (IgG) is only needed for low-abundance signals; multiple controlled studies indicate that when masking is absent, the coverage gain are modest and costs higher [28, 29]. Sputum should be liquefied and cleared of nucleic acids first; otherwise protein loss can reach 18% [30]. BALF and exhaled breath condensate (EBC) are characterized by low protein and high salt; the key is to concentrate first and then desalt, or downstream quantification will be overwhelmed by background ions [10, 31]. For extracellular vesicles (EVs), comparability across platforms remains limited; standardized isolation and quantification standards workflows are still needed [32]. Third, align purification with peptide class. Apply targeted enrichment only when needed; for modified peptides (e.g., phosphopeptides), immobilized metal affinity chromatography (IMAC) or metal oxide affinity chromatography (MOAC) is appropriate [33]. In brief, the matrix dictates the interference-removal strategy; the peptide class determines whether specialized enrichment is warranted; and the remaining steps should be harmonized in cleanup and acquisition to minimize between-batch variation. The pre-analytic framework is shown in Fig. 2, which summarises the stepwise clinical peptidomics workflow from sampling to data acquisition. Table 2 should be read alongside this figure; it maps each matrix–by–peptide-class combination to recommended enrichment and acquisition strategies and highlights the main risks and countermeasures. Pre-analytic quality control and decision rules for respiratory-sample peptidomics. (a) Quality-control workflow. Key checkpoints span the entire analytical chain: (i) rapid, cooled sampling and a ≤ 30 min window to solid-phase extraction (SPE); (ii) isotope and iRT spikes to track gradient drift; (iii) pooled-QC injections with CV < 20% and on-run retention-time charts; (iv) isotopic/FDR filters that hold the peptide-level false-discovery rate below 1%; (v) decision-curve and external, multicentre validation of prognostic models. Example chromatograms and a volcano plot illustrate retention-time stability and differential-abundance filtering. (b) Matrix- and peptide-class–guided preprocessing. The sample matrix dictates the first clean-up step: serum/plasma—Alb/IgG depletion if low-abundance masking is present; sputum—homogenise + DNase to counter viscosity/DNA; BALF/EBC—concentrate then desalt to correct low protein/high salt; urine—creatinine normalisation. A common clean-up (MWCO 3–10 kDa ultrafiltration + C8-SPE, RP) follows. The peptide class then decides on selective enrichment: Yes, PTM-oriented TiO₂/IMAC; No, charge-based SCX/WCX-SPE for endogenous fragments (ExMP). These choices stabilise downstream FDR and close the discovery-validation-translation loop Comparative analysis of key quality metrics Building on the workflow in Fig. 2, quantitative assessments at key steps show that enrichment choices drive identification gains: combining TiO₂ and Ti-IMAC captures more than twice as many phosphopeptides in a single pass while maintaining purity > 90% [34]. Miniaturization governs throughput and cost: an automated 96-well StageTip run on 96 plasma samples yielded a Pearson correlation of 0.98 across parallels, indicating cohort-scale stability [35]. Unspecific open searches require dual-level FDR control to suppress false positives: a single threshold lets false positives rise to 15%, whereas two tiers reduce them to 4% [36]. Overall, high-throughput endogenous peptidomics should prioritize optimized enrichment, batched miniaturized workflows, and dual-tier FDR control. Peptide class–by–matrix–by–technology matrix The workflow in Fig. 2 informs a matrix (Table 2) that maps the key steps and challenges for six peptide classes across different matrices. We first examine the matrix, then turn to the clinical interpretation that follows. Table 1Peptide classes across specimen matrices: pre-analytics, acquisition and QC essentialsPeptide classSerum/PlasmaSputumBALFEBCUrineKey refsPTM peptides (PTMP)Common: Protease inhibitors; desalting. Enrichment: TiO₂/IMAC (phospho), HILIC (glyco). Acquisition: DDA/DIA ± PASEF; PRM for targets. QC: site-level FLR/FDR ≤ 1% with localization[37–40]Extreme dynamic range; consider depletion or DIA‑PASEFHigh salt/viscosity; blanks/salt controls; residual proteasesLow biomass/adsorption; material/load controlTrace-level proteins; contamination monitoring firstpH/salt variability; CE‑MS or DIA/PRM feasibleProteolytic degradation fragments (PDF)Common: Rapid stabilization to prevent ex vivo cleavage. Workflow: 10 kDa MWCO →C18/SPE. Acquisition: DIA/directDIA for complex mixtures. QC: pooled QC, iRT, FDR ≤ 1%; infer protease–substrate links (Proteasix/MEROPS)[20, 21, 41]High-abundance proteins mask low-MW fragments; assess depletion impactActive proteases; add inhibitors; control viscosity (DTT ± DNase)Adsorption loss; use low-bind plastics; define blanksTrace signals; focus on contamination/ambient carryoverCE‑MS suits fragment fingerprints; robust reproducibilityCryptic peptides (CrP) Common: Peptides encrypted within parent proteins; released by proteolysis. Discovery via DDA/DIA; validate with PRM/MRM where bioactivity known [42, 43]High-abundance carriers; fractionation/desalting essentialECM-rich; watch exogenous protein/lipids; define blanksAdsorption to plastics; keep low-bind & pre-wettingUltra low content; contamination monitoring over depletionLow-MW rich; CE‑MS/DIA workableEndogenous bioactive peptides (EBioP) Common: Acidified low-temperature handling; SPE/C18 cleanup. Discovery by DDA/DIA; quantitation by PRM/MRM with SIL/AQUA standards [44–46]Matrix effects; prioritize targeted PRM/MRM with SIL.High salt/viscosity; desalting and blanks requiredLow biomass; concentration then cleanupAmounts at detection limit; method blanks mandatoryVariable pH/salt; CE‑MS or LC‑MS both feasibleExogenous/microbial peptides (ExMP)Common: Cleanup: low-salt, C18; enrichment via SCX/WCX‑SPE as needed. Identification: database + de novo; confirm by PRM/MRM for targets[47–49]Depletion or orthogonal fractionation may be neededHost/microbiome mix; DNase and viscosity controlLow load; enrichment before LC‑MS; blanks requiredTrace-level; method blanks & carryover checks are criticalLow-salt concentration then C18; WCX‑SPE optionalCirculating free peptides (CFP) Common: Desalting and cleanup (C18/StageTip); optional 10 kDa MWCO; Acquisition: DIA/directDIA for profiling; PRM for candidates [37, 39, 41]High-abundance proteins dominate; evaluate impact of depletionSalt/viscosity suppression; strict blanksLow concentration; adsorption controlUltra-trace; prioritize contamination monitoringStable platform for peptide fingerprints (CE‑MS/LFQ)Data-dependent acquisition (DDA), which fragments selected precursor ions to support confident identifications and spectral-library building. Parallel accumulation–serial fragmentation (PASEF), a timsTOF acquisition mode that increases sequencing speed and sensitivity. Other abbreviations are provided in the List of Abbreviations. ‘Common’ summarizes per-class baseline steps; matrix cells list only deltas. In-table [#] map to the reference list Peptide classes across specimen matrices: pre-analytics, acquisition and QC essentials Data-dependent acquisition (DDA), which fragments selected precursor ions to support confident identifications and spectral-library building. Parallel accumulation–serial fragmentation (PASEF), a timsTOF acquisition mode that increases sequencing speed and sensitivity. Other abbreviations are provided in the List of Abbreviations. ‘Common’ summarizes per-class baseline steps; matrix cells list only deltas. In-table [#] map to the reference list Each set of technology choices in Table 2 carries defined risks and corresponding countermeasures. Matrix-dependent differences in salinity and viscosity can impair quantitative accuracy; interference should be reduced through preconcentration, desalting, or viscosity-reducing steps [39, 40]. Targeted enrichment increases localization confidence but may reduce throughput; trade-offs should be weighed separately in the discovery and validation phases [46, 63]. For large discovery cohorts with limited sample volume, it is often preferable to prioritise simple, non-depleting cleanup with generic desalting to maximise throughput. By contrast, studies that aim to localise low-abundance or mechanistically specific peptide classes—such as HLA ligands, signalling peptides or microbial peptides—benefit from class-specific enrichment and stricter blanks, accepting reduced throughput and typically reserving such workflows for focused validation sets. Mass spectrometry platforms and acquisition modes High-throughput peptidomics now primarily uses electrospray ionization (ESI) coupled with LC-MS/MS and high-resolution mass analyzers (Orbitrap or quadrupole time-of-flight [Q-TOF]) [16, 50]. This section focuses on peptidomics-specific implementations of LC–MS/MS platforms, highlighting how endogenous peptides, often shorter and generated by non-specific proteolysis, impose different constraints than classical bottom-up proteomics [51]. Compared with tryptic peptides, endogenous peptides cover a wider range of m/z and charge states and often carry neo-termini or post-translational modifications, which lead to more heterogeneous chromatographic behaviour. Acquisition methods for clinical peptidomics therefore need careful optimisation of precursor selection, scan speed and dynamic range instead of simply copying proteomics defaults [51]. In DDA, the instrument cycles between survey scans and fragmentation scans (MS¹/MS²) on a subset of precursors that are selected in real time based on intensity and charge. It is conceptually simple and has underpinned many early immunopeptidomics and endogenous peptidomics studies, enabling high-confidence peptide identification and the construction of HLA allele- or tissue-specific spectral libraries [52, 53]. Its main limitation is stochastic precursor sampling. Low-abundance peptides in complex matrices may be selected in one run but not in another, which leads to missing values and complicates comparisons across large clinical cohorts. Ion-mobility–enabled DDA methods such as PASEF on trapped ion mobility spectrometry (TIMS) instruments increase sequencing speed and sensitivity, and can substantially expand HLA ligand coverage from limited input material [53]. In DIA, all precursors within predefined m/z windows are fragmented, producing multiplexed MS² spectra across the full mass range. Several immunopeptidomics studies have implemented SWATH-type DIA or diaPASEF to profile HLA ligandomes, reporting higher sensitivity, fewer missing values and better run-to-run reproducibility than high-resolution DDA when combined with experimental or predicted spectral libraries and heavy reference peptides [52, 54]. These applications benefit from constrained peptide length distributions (for example 8–11-mers for HLA-I) and well-defined binding motifs, which facilitate library generation and scoring. Applying DIA to complex biofluids for endogenous peptidomics is more difficult. Endogenous peptides span broader ranges of length, charge state and hydrophobicity, and overlapping degradation products from multiple proteases create dense, interfering fragment spectra in each DIA window [55]. Library-based identification is limited because only a fraction of all possible cleavage products can be sampled empirically, whereas purely library-free searches face very large search spaces and may yield higher false discovery rates if not carefully controlled. As a consequence, most DIA benchmarks have been carried out in proteomics and immunopeptidomics, and systematic evaluations in complex clinical peptidomes are still scarce [55]. In many recent clinical peptidomics and immunopeptidomics studies, discovery is increasingly based on DIA, followed by targeted validation using PRM or PRM-PASEF; however, empirical evidence in endogenous peptidomics remains more limited than in proteomics, and alternative workflows coexist [51, 54]. High-resolution DDA is still widely used for discovery in complex biofluids, and targeted SRM/MRM or PRM/PRM-PASEF assays built on small panels of endogenous peptides and stable isotope-labelled standards remain the mainstay for analytical validation and clinical translation [51]. Large scale tissue peptidomics combined with machine-learning models, such as cluster-based or bioactivity-prediction approaches, further illustrates how modern acquisition strategies can feed into computational frameworks that prioritise candidate bioactive peptides for functional testing [56, 57]. Database searching, DIA challenges, and FDR control in peptidomics For endogenous peptidomics, most workflows must accommodate semi-enzymatic or nonenzymatic cleavage patterns. Peptides frequently show semi-tryptic or nonspecific termini, and therefore database searches often combine unspecific or semi-tryptic enzyme settings with open or partially open modification searches [51]. Compared with strictly enzyme-constrained searches, these strategies recover noncanonical cleavage sites and unexpected post-translational or chemical modifications, but they enlarge the search space and make downstream scoring and false discovery rate (FDR) control more demanding [51]. Immunopeptidomics adds further constraints: search settings are typically no enzyme or very permissive, and results are filtered by peptide length (for example 8–11-mers for HLA class I) and by binding motifs or predicted HLA-binding scores to enrich for plausible ligands [58]. In DIA, these principles remain similar, but the search space also includes all fragment ions generated in each DIA window, so library design and matching scores must be adapted to the broader peptide space [59]. Several proteomics search engines are widely used in peptidomics. MaxQuant/Andromeda, MSFragger, PEAKS and related tools can all be configured for semi-tryptic, nonspecific or “no-enzyme” searches with extended variable-modification lists [51]. For DIA data, software such as DIA-NN and Spectronaut supports both library-based and library-free analysis and can use predicted spectra to score noncanonical peptides [59], and recent dia-PASEF and data-independent immunopeptidomics studies illustrate how these tools can be applied to low-abundance peptidomes. Dedicated immunopeptidomics pipelines, such as MHCquant and similar frameworks, integrate “no-enzyme” searches with peptide-length and motif filters, HLA-binding prediction and FDR estimation to produce class I and class II ligand lists with controlled error rates [58]. As summarised in Table 2, nonenzymatic workflows and open searches entail higher error rates than strictly enzyme-constrained searches and therefore require thresholds at multiple levels together with orthogonal validation [51]. In peptidomics, FDR is typically controlled at the PSM and peptide levels, but additional safeguards are often used: peptide-length– or motif-stratified FDR thresholds, entrapment strategies using sequences from non-target species, and checks for concordance with other omics layers or independent cohorts [51, 57]. In immunopeptidomics, motif-based filters and HLA-binding predictions act as an extra layer of error control by removing peptides that do not fit allele-specific length or motif constraints [58]. DIA adds further challenges because fragment ions from many precursors are multiplexed in each window; best practice is to report FDR at the precursor or peak-group level as well as at the peptide level and to document library construction and scoring models in detail [59]. Recent work on peptide clustering has shown that grouping related peptides into clusters can sharpen protease-cut signatures and improve large-scale statistical analyses in wound-fluid and clinical peptidomes [57]. Quality control and data sharing Quality control (QC) is essential for translating high-throughput peptidomics into clinical practice. Early plasma DIA studies relied on indexed retention time (iRT) calibration to correct gradient drift, I but issues with pump-pressure fluctuations and ion-source contamination persisted [60]. Subsequently, laboratories paired iRT calibration with a pooled quality-control sample (QC pool) to monitor within- and between-batch CVs, identification counts, and data completeness, thereby flagging reruns or triggering instrument maintenance when needed [61]. Because endogenous peptides are shorter and often lower in abundance than tryptic peptides, variability in CVs, iRT alignment and ID numbers can be more pronounced than in proteomics, making QC pools and isotopically labelled spike-in standards particularly critical in clinical peptidomics. DIA commonly uses the PQ500 stable isotope peptide reference mix for quantitative accuracy, which is helpful for designing targeted peptidomics assays. At the targeted validation stage, MRM/PRM strategies incorporating AQUA or similar internal standards typically keep CVs for most peptides below 20%, meeting reproducibility requirements for validation and translational research [62]. Despite these advances, true large-scale and multicentre clinical peptidomics cohorts are still rare, so harmonised QC frameworks and open data are essential for comparability across studies and centres. For data sharing, community resources provide the foundation for reanalysis and method benchmarking. Early large scale neuropeptidomics studies already demonstrated that depositing well-annotated peptidomics datasets in repositories such as PRIDE enables secondary analyses and cross-cohort comparisons [63]. Resources such as ProteomeXchange, a federated portal, had indexed 45,801 peptidomics datasets as of 13 September 2025 [64]. The Human PeptideAtlas January 2025 release (v2025-01) integrates 837 datasets [65]. These shared resources provide a reproducible foundation for algorithm benchmarking, interlaboratory QC threshold setting, and future pooled analyses across large cohorts. Data integration and model development Quantification, sparsity, and downstream statistical analysis In respiratory peptidomics, raw spectra are first reduced to peptide-level quantitative readouts. Discovery workflows based on DDA or DIA usually rely on label-free intensity measures such as integrated peak areas or MS¹ precursor intensities, whereas targeted SRM/MRM and PRM/PRM-PASEF assays quantify peak areas or area ratios to stable-isotope-labelled standards for relative or absolute measurement [66]. Spectral-count–based quantification still appears in some studies but generally offers lower dynamic range and precision than intensity-based approaches [66]. Across platforms, simple normalization (for example global scaling, median or quantile normalization, or the use of internal standards) and log transformation are needed to control batch effects and make samples comparable [67]. Peptidomic data matrices are typically sparse: many low-abundance peptides are only detected in a subset of samples. It is therefore common to remove features with very low detection rates before statistical testing to reduce noise and the multiple-testing burden [67]. A key step is to distinguish technical missingness (values censored below the limit of detection or quantification) from biological absence [66]. For left-censored values, small-value substitution or dedicated imputation schemes can be used, whereas values that are plausibly biologically absent are often left as missing and handled directly by the statistical model [66, 67]. Linear-model and generalized-linear-model frameworks, including limma-type moderated linear models and edgeR-like approaches, provide flexible ways to analyse such high-dimensional data while borrowing strength across peptides [68, 69]. For downstream analysis, single-peptide differential abundance is usually assessed with linear-model frameworks when assumptions are approximately met, and with rank-based tests or generalized models for skewed or zero-inflated data. Multiple-testing correction and the reporting of effect sizes with confidence intervals are essential to judge the strength and consistency of evidence [68]. Beyond univariate testing, pathway and network methods can combine multiple peptides per protein, pathway or tissue compartment. Degradomics-style analyses of cleavage sites and protease–substrate networks, supported by resources such as Proteasix and MEROPS, help infer active proteases and prioritize peptides that are most likely to be biologically meaningful candidates for targeted validation [20, 70]. AI-based de Novo peptide sequencing Prior to inference, tandem spectra undergo standard preprocessing such as peak picking, deisotoping, charge deconvolution and mass calibration [71, 72]. DeepNovo first combined convolutional neural networks and long short-term memory (CNN and LSTM) f architectures and substantially improved peptide-level accuracy over earlier tools [73]. pNovo3 then introduced a learning-to-rank strategy to rerank close-scoring candidates, with particular gains for short peptides [74]. Transformer-based models such as Casanovo reframed de novo sequencing as sequence-to-sequence prediction and improved cross-species performance [75]. Most recently, semi- or non-autoregressive Transformer architectures exemplified by TSARseqNovo have increased recall and speed for large DIA datasets, and public benchmarks such as NovoBoard now provide a unified basis for comparing these algorithms across species and instruments [76, 77]. The main value of AI-based de novo sequencing in peptidomics is biological rather than algorithmic. It can recover peptides outside existing databases and spectral libraries—including immunopeptides, cryptic or truncated peptides, sequence-variant peptides and PTM-bearing peptides—that are relevant to host–pathogen interactions and tissue-specific proteolysis [42, 43]. In practice, de novo tools are best used as discovery engines that generate high-confidence candidate lists, which are then confirmed by targeted PRM/MRM assays or synthetic reference peptides in well-phenotyped cohorts before being considered robust biomarkers. Machine learning for feature selection and biomarker discovery In peptide biomarker studies, basic statistics, such as univariate tests and regression, are the first step and work well for single peptides or small panels, providing effect sizes and simple calibration. Their ability to handle high dimensionality, strong correlations and non-linear relationships is limited, especially when hundreds of peptides are analysed together with clinical and imaging variables. In these settings, machine-learning (ML) methods become useful to model complex patterns while still aiming for compact, clinically usable panels. During biomarker selection, batch effects and confounding variables should be adjusted before model development. Feature selection options include sparse regression methods such as least absolute shrinkage and selection operator (LASSO) and elastic net, as well as minimum redundancy maximum relevance (mRMR) and the Boruta algorithm [78–82]. Nested cross-validation is recommended to prevent information leakage when tuning hyperparameters or comparing several classifiers, and deep learning can be considered when relationships among predictors are highly non-linear, ideally with pathway information incorporated to enhance interpretability [83]. Development and validation of Diagnostic/Prognostic models In clinical translation, peptide panels often outperform individual peptides. The CKD273 urinary peptide classifier is a useful template: differential urinary peptides were identified in a discovery cohort, replicated across independent populations, refined into a multipeptide panel and then taken forward towards regulatory engagement for risk stratification in chronic kidney disease [84]. When evaluating diagnostic or prognostic models, discrimination metrics such as the area under the curve (AUC), sensitivity and specificity are necessary but not sufficient. Calibration, defined as the agreement between predicted risks and observed event rates, should also be reported, for example using calibration plots or the Brier score [85]. A simple counterexample is a model that classifies almost everyone as “high risk”: the AUC can still look acceptable, but predicted probabilities are clearly misaligned with reality, indicating poor calibration. TRIPOD-AI provides a reporting guideline for studies that develop or validate prediction models using regression or machine learning, while PROBAST-AI extends the PROBAST tool to assess risk of bias and applicability of such models [86, 87]. Adhering to these frameworks can reduce overfitting and selective reporting, and improve the transparency, comparability and reproducibility of future peptide-based prediction models. Applications of peptidomics in respiratory diseases Over the last five years, applications of peptidomics to respiratory diseases have yielded two prototypical classes of biomarkers. First, fragment peptides generated by protease imbalance in the airway or biofluids can serve as indicators of disease activity or prognosis—for example, the C-terminal peptide of alpha-1 antitrypsin (AAT) and antimicrobial peptides [88]. Second, fibrosis-related extracellular matrix (ECM) neo-epitope peptides are associated with disease stage and progression [89]. Compared with lung cancer, clinical peptidomics studies in COPD and asthma remain relatively scarce. Even so, mass spectrometry and bioinformatics workflows using EBC, exhaled breath particles (EBP), and serum as matrices have established a reusable framework for analysis across samples. Figure 3 provides a visual overview of these recent advances, mapping the convergence of diseases, analytical methods, and sample matrices in respiratory-disease peptidomics. The following sections focuses on COPD, while also detailing representative findings and methodological highlights for asthma, lung cancer, pulmonary fibrosis, other respiratory disorders. Respiratory-disease peptidomics landscape (last 5 years). A tri-partite diagram maps recent peptide discoveries in respiratory medicine. Top (pink)—six lung diseases with reproducible peptide signals: COPD, asthma, lung cancer, pulmonary fibrosis, bronchiectasis and COVID-19. Left (blue)—six modular methods driving those studies: LC-MS instrumentation, DIA/SWATH acquisition, SPE, 3–10 kDa MWCO filtration, PTM-enrichment by IMAC/TiO₂, and charge-based SCX/WCX fractionation. Right (green)—five sample matrices that support cross-cohort replication: exhaled-breath condensate/particles (EBC-EBP), sputum, BALF, serum and urine. The central hub underscores the converging potential of matrix-appropriate methods and disease-specific peptide panels for clinical translation COPD This section synthesizes COPD-specific differential peptidomic profiles, assesses evidence consistency and how methodological differences affect conclusions, and discusses their potential for clinical translation. Key differential peptidomic profiles and functions For ease of exposition, the following sections organize recurrent differential peptide signals by functional categories reported in recent years. Initially, peptides linked to proteolysis and extracellular matrix (ECM) remodeling tend to be elevated across multiple studies; for example, elastin- and collagen-derived fragments have been reported in both sputum and serum and can distinguish frequent exacerbators, showing some cross-matrix consistency and corresponding to the treatable trait of excessive ECM turnover, although the evidence still stems mainly from single-center observational cohorts with limited external replication. Among urinary markers, total desmosine (tDES) is positively associated with long-term mortality, but it is influenced by smoking status and renal function; stratification and covariate adjustment are therefore required in statistical modeling [90, 91], and current data again come primarily from a few single-center cohort studies rather than large multicenter validation. Subsequently, representative mucosal-defense peptides—such as secretory leukocyte protease inhibitor (SLPI) and β-defensin-1—show significant inverse correlations between their sputum concentrations and FEV1 in patients in the stable phase, suggesting that they more directly reflect airway inflammatory activity and tissue remodeling [92], and map onto the treatable traits of neutrophilic airway inflammation and impaired mucosal defense, with support so far limited to cross-sectional single-center studies. Most related studies focus on proximal airway biofluids, such as sputum and BALF. Owing to their proximity to the site of pathology, these matrices exhibit higher detection rates and better consistency, but prospective external replication is scarce. Ultimately, peptides linked to inflammation and complement activation are elevated in patients with frequent exacerbations, supporting their potential use in risk stratification and reflecting systemic-inflammation and exacerbation-prone treatable traits; however, external validation remains limited [93]. Beyond these signals, higher levels of oxidatively modified antiproteases have been observed in parallel with oxidative stress markers and disease severity, suggesting involvement of a protease–inhibitor imbalance; nevertheless, evidence remains sparse and warrants prospective validation [94]. Strength and consistency of evidence In recent years, studies of COPD-related peptide or protein signals have largely been single-center and cross-sectional or retrospective in design, with per-study sample sizes typically in the tens to a few hundred. Only a subset has performed targeted confirmation by PRM/MRM or external validation in an independent cohort; prospective and multicenter validation remains limited, so the overall strength of evidence is moderate [95]. With respect to direction and consistency, mucosal defense peptides such as secretory leukocyte protease inhibitor (SLPI) and β-defensin-1 show inverse associations between their sputum concentrations and forced expiratory volume in 1 s (FEV1) in patients in the stable phase, and these findings have been replicated in independent studies; however, reproducibility during exacerbations remains limited, so these peptide signals are still hypothesis-generating [92]. In serum-based peptidomics, inflammation- and complement-driven peptides have generally undergone internal confirmation and can distinguish frequent exacerbators, yet independent external validation cohorts are largely lacking, indicating a need for prospective validation [96, 97]. Impact of methodological differences on conclusions In COPD research, certain study specific practices can affect the robustness of conclusions. Regarding acquisition and library strategy, many recent studies employ DIA with predicted or public spectral libraries, while others continue to apply DIA with project specific empirical libraries, so these workflows currently coexist. Differences in library content, retention time alignment, instrument platform and scoring thresholds between centres can lead to partially distinct sets of peptides passing false discovery rate control, which reduces the apparent overlap of candidate signatures across cohorts. Transparent reporting of library source and version, PRM/MRM confirmation of key peptides, and reassessment of key metrics in an independent cohort increase the likelihood of robust findings [98–100]. Details of pre-analytical processing and enrichment are provided under “Pre-analytical Processing and Enrichment Strategies”; here we focus on their impact on results. Alternative workflows can alter the repertoire and relative abundances of detectable peptides, thereby affecting the direction of effects and reproducibility. Common practice is to include blank controls and spike-in experiments to document recovery and the lower limit of quantification (LLOQ) and to perform sensitivity analyses [101]. For low-abundance matrices, the choice of preconcentration and normalization often shapes reproducibility; transparent reporting of these strategies and reproducibility metrics facilitates comparison across studies [102]. On quality control, the field now largely follows the framework outlined under “Quality Control and Data Sharing”. For statistics and validation, preferred practices include prespecifying thresholds, retaining an independent test set, and performing external validation in an independent cohort [103]. Overall, these points help explain the inconsistencies in the functional signals reported across studies and offer actionable directions for subsequent validation and clinical translation. Clinical translation and evidence gaps Most COPD-related peptidomics studies remain at an early stage of the validation pathway. In line with the decision matrix in Fig. 1/Table 2 and the FDA–NIH BEST framework, translation should follow four steps [104]. Discovery: DIA-based peptidomics or shotgun proteomics in well-phenotyped cohorts define candidate panels for a prespecified intended use, with adjustment for batch effects and confounders. Technical validation: targeted LC–MS/MS (PRM/MRM with stable-isotope–labeled peptides) or ELISA establishes linearity, precision, LLOQ, and stability. Biological/clinical validation: in independent cohorts, multivariable models test incremental discrimination and calibration, and decision-curve analysis evaluates clinical benefit. Clinical implementation: harmonized cut-offs, external quality assessment, and pragmatic studies are required before routine use. A serum-based 123-peptide SRM panel distinguishing chronic airflow limitation from COPD illustrates this pathway but still lacks multicenter trials demonstrating utility [17]. Overall, the main gaps are insufficient targeted validation, limited external and prospective cohorts, and non-standardized assays and decision-analytic evaluation [45]. A structured summary of key COPD related peptide signatures, their sample types, clinical relevance, and validation stages is provided in Table 3. Other respiratory diseases Table 5 summarizes peptidomics evidence across respiratory diseases, organizing studies by disease, specimen and disease state, analytical workflow, representative peptide or protein signatures, reported performance, and overall evidence level. Beyond COPD, studies over the last five years have identified reproducible peptide or protein signals in asthma, lung cancer, pulmonary fibrosis, bronchiectasis, and pulmonary hypertension, suggesting potential utility for disease subtyping, early diagnosis, and prognostic assessment. In asthma, mass spectrometry studies of EBC have identified molecular signals that distinguish severe from mild-to-moderate disease; serum and induced sputum studies likewise differentiate inflammatory phenotypes, consistent with airway inflammatory activity [109, 110]. In lung cancer, low-molecular-weight peptide profiles in serum or plasma can distinguish patients from controls, and several studies have validated these findings in independent cohorts. Plasma-based proteomic signatures can also differentiate lung adenocarcinoma from COPD [111, 112]. In pulmonary fibrosis, circulating proteomic signatures correlate with survival risk, suggesting utility for risk stratification and treatment-response evaluation [113]. In bronchiectasis, sputum proteomic profiles reveal distinct exacerbation subtypes, with signals associated with Pseudomonas aeruginosa being the most pronounced [114]. In pulmonary arterial hypertension, plasma proteomic signatures stratify patients into groups independently of clinical classification, indicating potential for patient stratification and assessment of therapeutic response [115]. Compared with COPD, where sputum- and serum-based signatures have already entered early technical and clinical validation with small independent cohorts, most other respiratory diseases still rely mainly on single-center exploratory studies with modest sample sizes and scarce external replication, and therefore remain earlier on the discovery–validation continuum. Table 3Overview of human peptidomics evidence across respiratory diseases (2019–2025)DiseaseSpecimen & stateWorkflowRep. peptides/sig.Dir.UsePerformanceEvidence; refsCOPD*Sputum; stable COPD vs. asthma/HCMSB + MALDI pept.α‑defensins (HNP1‑3) ↑; AMP peptides↑Pheno (discovery)N/Aretrospective discovery; single‑center; ext N; refs [106, 116]Asthma*Airway secretions; allergic vs. HC/vs. COPDMSB‑MALDI; nasal immunoassayHNP1‑3, LCN2 ↓ in nasal; sputum less neutrophil‑dominant↓ (HNP1‑3)Endotyping (discovery)N/Aretrospective discovery; single‑center; ext N; refs [117, 118]NSCLCSerum; NSCLC vs. HCMB‑WCX + MALDI pept.FGA fragments ↑↑Dx (disc.+int. val.)Se 92.9%; Sp 91.7%retrospective; single‑center; ext N; refs [111]TBSerum/plasma; active TB vs. HC/othersHost peptidomics scarce; LC‑MS/MS proteomicsComplement/coagulation/APP proteins; endogenous peptides not establishedN/ADisc./Dx researchN/Acontext proteomics; multicenter; ext N/A; refs [119]NTM‑PDCulture isolates (resp.)MALDI‑TOF protein ID; nucleotide MALDI‑TOF— (pathogen ID, not host)N/ASpecies ID (routine)Varies; multicenter reproduciblemethodology; multicenter; ext N/A; refs [120–123]ARDS*BALF; pneumonia‑ALI/ARDSHydrophobic frac. + MALDI peptide patternsPattern signatures (no single peptide)—Prog (discovery)Model perf. reportedretrospective discovery; single‑center; ext N; refs [124]IPA*Plasma/serum; hematologic/HSCTSID‑SRM host peptides + GM/BDG antigensComplement/coagulation/APP peptide panel↑/↓ (panel)Dx (panel)Panel > GM aloneretrospective; single‑center; verification; ext N; refs [125]BronchiectasisSputum/BALF; stable/exac. endotypesLabel‑free proteomics; immunoassaysLL‑37↑; SLPI↓; NE activity↑ ~ severity/exac.↑/↓Endotyping; risk strat.N/Aobservational; single‑center; ext N; refs [126, 127]COVID‑19Plasma; mild–severe vs. HCLC‑MS/MS endogenous pept.FGA‑derived peptides ↑ with severity; APOA1 changes↑/↓Severity (discovery)N/Aretrospective discovery; single‑center; ext N; refs [128]COVID‑19Serum; Dx/SeverityMALDI fingerprint + MLSpectral features (acute‑phase/AMP regions)—Dx/Prog (disc.; int. vals.)Acc. 90–99% (Dx)multiple cohorts; mostly internal val.; ext N; refs [129, 130]PAH————No qualifying host peptidomics—proteomics rich; ext N/A; refs [131, 132]IPF————No clinical host peptidomics—proteomics rich; ext N/A; refs [133, 134]HIV‑OLDBALF & plasma; PLWH ± OLDEndogenous peptidomics + proteomicsPeptides map to NE/cathepsins/granzymes; assoc. FEV1%—Endotyping; pathways—observational; single‑center; ext N; refs [135]Direction of change (Dir.): ↑ increase, ↓ decrease, — no change, ↑/↓ mixed. Performance metrics: area under the curve (AUC, preferred; 95% CI), otherwise sensitivity (Se), specificity (Sp) or accuracy. Evidence = design; center; ext (Y/N/N/A); refs [#]. *pre-2019; N/A not applicable; NR not reported Overview of human peptidomics evidence across respiratory diseases (2019–2025) Direction of change (Dir.): ↑ increase, ↓ decrease, — no change, ↑/↓ mixed. Performance metrics: area under the curve (AUC, preferred; 95% CI), otherwise sensitivity (Se), specificity (Sp) or accuracy. Evidence = design; center; ext (Y/N/N/A); refs [#]. *pre-2019; N/A not applicable; NR not reported From peptidomics to treatable traits and peptide therapeutics Mapping to treatable traits Peptidomic signatures in COPD can be grouped into treatable traits that correspond to distinct biological axes and link molecular findings to risk based and response based management. In this section we highlight three examples that are already supported by the evidence summarised in previous disease specific Sect [12]. Elastin and collagen derived fragments, such as urinary tDES, reflect accelerated extracellular matrix turnover and could help identify emphysema predominant or rapidly progressive patients and monitor extracellular-matrix turnover under anti protease or anti fibrotic therapy. However, cut off values and intra individual variability are not standardised, most peptide panels have not been validated in external multicentre cohorts, and integration with imaging and genetic risk scores remains limited [90]. Sputum antimicrobial peptides such as SLPI, β-defensin-1 and α-defensins define a neutrophilic inflammation/mucosal-defence trait that may support risk stratification of chronic-bronchitis and frequent-exacerbator phenotypes and monitoring of anti-neutrophilic or macrolide-based regimens. Assays are not harmonised across matrices, peptide levels are influenced by inhaled therapies and sampling techniques, and robust external validation or multi-omics integration linking these signatures to specific treatment decisions is still scarce [92]. Complement derived activation peptides in plasma represent a complement and systemic inflammation trait that links airway injury to systemic comorbidities and can contribute to exacerbation risk stratification and pharmacodynamic assessment of anti inflammatory or complement targeted therapies [93]. Available data are largely observational and single centre, analytical workflows and sampling strategies are heterogeneous, and no trial has yet tested treatment guided by complement related peptide panels or large scale multi omics integration [12]. Peptide therapeutics Three overarching trends characterize the development of peptide therapeutics. First, the global development pipeline continues to expand, with agents targeting metabolic and cardiovascular indications remaining predominant [136–138]. Representative advances include glucagon-like peptide-1 (GLP-1) receptor agonists and newer multi-receptor co-agonist peptides, which show clinical value for weight management and for cardiorenal outcomes [138–141]. For respiratory diseases, these drugs are relevant to systemic treatable traits such as obesity, metabolic syndrome and cardiovascular comorbidity that strongly modify COPD and asthma outcomes. Second, half-life engineering and delivery approaches have advanced substantially. Lipidation, conjugation to albumin or immunoglobulin fragments, cyclization, and terminal modifications are used to prolong in vivo half-life. In parallel, sustained release microspheres and implantable formulations, together with exploration of noninjectable routes such as oral, intranasal, and inhalation delivery, have broadened the range of applications [142–145]. For chronic airway diseases that require long term maintenance therapy and prevention of exacerbations, these technologies can reduce injection burden, support adherence and enable high local drug concentrations in the lung. Third, immunology and infectious disease are seeing renewed momentum. Antimicrobial peptides and immunomodulatory peptides have entered early clinical evaluation, and exploratory efforts now target respiratory indications, including inflammatory control in COPD and the management of drug-resistant infections [146–148]. These programmes align with treatable traits such as neutrophilic inflammation, impaired mucosal defence and recurrent infection, and illustrate how peptide drugs can be designed along the biological axes identified by respiratory peptidomics. Together, these developments outline a translational pathway from peptidomics to peptide based interventions. Peptidomic studies can identify active pathways and candidate targets within defined treatable traits; medicinal chemistry and half life engineering can convert these into optimised peptide or peptide inspired molecules; and early phase respiratory trials, including inhaled and systemic peptides, provide a setting in which peptidomic biomarker panels can serve as pharmacodynamic readouts and tools for patient selection. Conclusions and future directions Within the sampling, preparation, and quality control framework outlined above, the most pressing challenges are peptide loss during pre-analytical processing and enrichment, and identification bias arising from mismatches between DIA data and spectral libraries; these factors amplify batch effects and undermine reproducibility. On the sample side, confounders such as smoking and renal function can affect urinary peptide signals. Elastin-derived biomarkers have been associated with long-term mortality in COPD, but rigorous stratification is required. Future work can leverage single-cell and spatial omics to localize the cellular origins of peptides and to map local protease activity. In parallel, federated learning (FL) can be used to update models across centers while preserving privacy, thereby accelerating the clinical translation of peptide panels. Overall, across specimen types and analytical platforms, endogenous short peptides consistently indicate extracellular matrix remodeling and inflammatory activity, and show potential amenable to validation for COPD phenotyping, risk stratification of acute exacerbations, and early diagnosis. The main limitations currently stem from pre-analytical variability, batch effects, and inconsistencies in spectral libraries, which undermine cross-center comparability and constrain clinical interpretation. Evidence is most mature in COPD, whereas other respiratory indications remain largely exploratory; therefore, priorities for clinical translation include standardized pre-analytics and QC, multicenter external validation, multimodal integration with other omics and clinical outcomes, and community-shared spectral libraries with deposition of raw data. In parallel, investigators should establish open spectral libraries and raw data repositories, explore noninvasive specimens such as EBC, and advance spatial and single-cell peptide measurements to resolve the lesion microenvironment. We further recommend federated modeling that preserves privacy to enable model updating across centers without sharing raw data. Taken together, these approaches may shorten the path from discovery to clinical use for peptide biomarkers, with initial implementation anticipated in COPD and subsequent extension to asthma, lung cancer, and pulmonary fibrosis. Acknowledgements Not applicable. Abbreviations Alpha-1 antitrypsin Artificial intelligence Albumin Absolute QUAntification Area under the curve Bronchoalveolar lavage fluid Biomarkers, Endpoints, and Other Tools (FDA–NIH BEST framework) Urinary peptide classifier comprising 273 peptides for chronic kidney disease Chronic obstructive pulmonary disease Creatinine Coefficient of variation Diabetic kidney disease Data-dependent acquisition Data-independent acquisition DIA analysis software based on neural networks Data integration analysis for biomarker discovery using latent components Deoxyribonucleic acid Exhaled breath condensate Exhaled breath particles Extracellular matrix Extracellular vesicles Exogenous/microbial peptides U.S. Food and drug administration False discovery rate Forced expiratory volume in 1 s Federated learning Global burden of disease Glucagon-like peptide-1 Immunoglobulin G Immobilized metal affinity chromatography Indexed retention time Kilodalton Least absolute shrinkage and selection operator Liquid chromatography Liquid chromatography–tandem mass spectrometry Lower limit of quantification Matrix-assisted laser desorption/ionization time-of-flight Metal oxide affinity chromatography Multi-omics factor analysis Multiple reaction monitoring Minimum redundancy maximum relevance Mass-to-charge ratio Mass spectrometry First-stage (precursor) / second-stage (fragment) mass spectra Molecular weight cutoff National institutes of health Parallel accumulation and serial fragmentation Prediction model risk of bias assessment tool – Artificial Intelligence Parallel reaction monitoring PRM combined with PASEF Post-translational modification(s) ProteomeXchange Quadrupole time-of-flight Quality control Pooled quality-control sample NCBI Reference sequence database Reversed phase Strong cation exchange Surface-enhanced laser desorption/ionization time-of-flight Secretory leukocyte protease inhibitor Solid-phase extraction Selected reaction monitoring Sequential window acquisition of all theoretical fragment ion spectra Titanium dioxide Trapped ion mobility spectrometry Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis – artificial intelligence Total desmosine Weak cation exchange Years lived with disability Author contributions QingYu Zhou (Q.Y.Z.) conceived the study, collected and analysed the data, and drafted the manuscript. YaHui Shen (Y.H.S.) designed and supervised the study and performed critical revision of the manuscript for important intellectual content. Both authors read and approved the final manuscript. Funding This research was supported by the Taizhou Science and Technology Support Program (grant TS202412). The funder had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript. Data availability All data generated or analysed during this review are included in this published article (Table 1, and 2). No new raw datasets were generated. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated or analysed during this review are included in this published article (Table 1, and 2). No new raw datasets were generated.