Residue‑resolved dynamically averaged interaction analysis of direct factor Xa inhibitors by MD‑FMO combination calculations Authors: Ryohei Yoshine, Yoshinori Hirano, Hideo Doi, Shun Kitahara, Kazuki Akisawa, Koji Okuwaki, Eiji Yamamoto, Kenji Yasuoka, Yuji Mochizuki Published: 2026 ABSTRACT Four widely used direct oral anticoagulants (DOACs)—Edoxaban, Betrixaban, Apixaban, and Rivaroxaban—inhibit factor Xa, a key enzyme in the blood coagulation cascade. Despite their pharmacological importance, theoretical analyses of their binding modes with factor Xa remain limited. In this study, we performed fragment molecular orbital (FMO) calculations to quantitatively analyze inhibitor-protein interactions. Structural ensembles were obtained from molecular dynamics (MD) simulations employing classical force fields. Interaction energies between each ligand and the amino acid residues in the binding pocket region were dynamically averaged over MD-derived structural fluctuations. Comparative analyses revealed both common and distinct binding features among the four inhibitors at residue-level resolution. As a result, Cys191, Tyr99, Trp215, Glu217, Phe174, and Gln192 were identified as major contributive residues across the four ligands, with cooperative stabilization arising from electrostatic, charge-transfer, and dispersion components. The influence of protonation (charged states) was also examined for Edoxaban and Betrixaban.Supplementary InformationThe online version contains supplementary material available at 10.1007/s10822-026-00846-x. Introduction Cardiovascular diseases, including myocardial infarction and stroke, are among the leading causes of mortality worldwide, and thrombosis plays a central role in these conditions. The formation of blood clots in arteries and veins is known to cause heart attacks and strokes [1–3]. Therefore, various anticoagulant drugs that inhibit the blood coagulation (or thrombosis) cascade, in which numerous proteins are involved [4], have long been used [5–17]. Although Warfarin has been used for a long time, its food interactions and potentially high bleeding risk have led to the more widespread use of so-called direct oral anticoagulants (DOACs), most of which inhibit (or ban) coagulation factor Xa [6–11]. During the 2020–2022 SARS-CoV-2 pandemic, thrombosis emerged as a major concern, leading to the exploratory use of DOACs in clinical practice and clinical trials [18–21]. Among DOACs, Edoxaban [22], Betrixaban [23], Apixaban [24], and Rivaroxaban [25] are currently widely used; however, regional differences in their use exist, and therefore the respective brand names are omitted here. A detailed molecular-level understanding of these four DOACs is crucial, as it can guide the optimization and development of next-generation anticoagulant drugs. Among computational chemistry studies on anticoagulants, quantitative structure–activity relationship (QSAR) approaches [26–30], including molecular docking techniques [31], are among the first to be mentioned. In simulation-based research, several studies employing Monte Carlo (MC) [32] and molecular dynamics (MD) methods [33–42] have been reported. Analyses using molecular orbital (MO) calculations [43–45] and systems biology-based approach [46] have also been reported. Several MD‑based studies are particularly relevant here. Abel et al. evaluated binding free‑energy changes for numerous factor Xa ligands with an emphasis on hydration effects and identified Ser195 as a key residue contributing to ligand recognition and affinity [34]. Du et al. investigated Edoxaban, Betrixaban, and related ligands, estimating their binding free energies and providing residue‑specific views of their interactions [37]. While these studies are highly informative, even more detailed insight into hydrogen-bonding, electrostatic interactions, and dispersion can be obtained by employing non-empirical MO methods when feasible. The fragment molecular orbital (FMO) method, originally proposed by Kitaura [47], is now one of the most widely adopted fragment-based approaches [48–51] in biophysical and pharmaceutical chemistry [52, 53]. Extensive developments in the theory and applications of FMO have been systematically summarized in a couple of review articles [54, 55]. The popularity of FMO stems mainly from two advantages. First, the method exhibits excellent scalability on parallel computing architectures, enabling routine calculations on proteins containing hundreds to thousands of amino acid residues [53, 55]. Second, as a non-empirical quantum–mechanical method, the FMO approach provides practical quantitative measures for dissecting inter-fragment interactions in complex molecular systems. These interaction indices are commonly known as inter-fragment interaction energy (IFIE) [56], pair interaction energy (PIE) [57], and their component-decomposed form, pair interaction energy decomposition analysis (PIEDA) [57–59]. Applications of the FMO method in the pharmaceutical field have also been collected in Ref. [60]. Until the 2010s, aside from a few pioneering studies [61, 62], most drug-related FMO interaction analyses [53, 55, 60] were restricted to single static structures taken from the protein data bank (PDB). Since the 2020s, the availability of high‑performance supercomputers exemplified by Fugaku has enabled routine workflows in which droplet models are extracted from (classical) molecular dynamics (MD) trajectories and subjected to batches of FMO calculations for dynamically averaged or statistical analyses of molecular interactions [63–70]. The significance of this MD-FMO combination framework, which better reflects realistic biochemical conditions, has been highlighted in Ref. [71]. In this paper, we report a series of dynamically averaged interaction analyses on complexes between factor Xa and four DOACs: Edoxaban [22], Betrixaban [23], Apixaban [24], and Rivaroxaban [25]. These statistical analyses are based on combined MD-FMO calculations. Particular attention is paid to clarifying the contributions of residues in the binding pockets, as well as to characterizing similarities and differences among the four drugs. Method of calculation Preparation of fundamental structure The upper part of Fig. 1 shows the chemical structures of four DOACs (in their neutral forms), while a schematic coagulation cascade by DOAC is provided in Fig. S-1 of the Supporting Information (SI) for reference. The lower part of Fig. 1 illustrates the four binding pockets of factor Xa (S1–S4) [12, 34, 37]. The S1 and S4 pockets are considered the primary sites of DOAC binding, and Edoxaban is shown as a representative ligand to illustrate this binding mode. In particular, the S4 pocket is known to play a crucial role by providing hydrophobic and stacking interactions [23, 37], and thus dispersion-type stabilization is expected to be substantial. Following our previous study [37], docking of these four ligands to factor Xa was performed using the molecular modeling software MOE [72]. The crystal structure with PDB ID 2EI8 [73] was used as the template. The protonation states of amino acid residues in factor Xa were assigned at pH 7, 300 K, and a salt concentration of 0.1 M using the Protonate 3D function in MOE. The two structural Ca2+ ions present in the 2EI8 structure were retained for the MD simulations.Fig. 1Chemical structures of four DOACs (upper) and four binding pockets of factor Xa (lower). Edoxaban is shown as a representative ligand to illustrate the binding to the S1 and S4 pockets Chemical structures of four DOACs (upper) and four binding pockets of factor Xa (lower). Edoxaban is shown as a representative ligand to illustrate the binding to the S1 and S4 pockets For Edoxaban [22], two structures with different charge states were prepared to account for pH-dependent protonation in vivo. Edoxaban(N) in Fig. S-2 left of SI is the neutral form, whereas Edoxaban(N+) (right) corresponds to the protonated form (at pH 7) in which a positive charge is located on the nitrogen atom of the thiazolopyridine scaffold (indicated by the blue arrow). In the same manner, two structures with neutral and protonated states were prepared for Betrixaban [23]. Betrixaban(N) in Fig. S-3 left is the neutral form, whereas Betrixaban(N+) (right) carries a positive charge on the nitrogen atom of the dimethylbenzimidamide scaffold due to protonation. For Apixaban [24], two structures, Apixaban(S1) and Apixaban(S4), were considered to account for possible differences in modes to the binding pockets. However, Apixaban moiety moved away from the S1 pocket during the early stages of the MD simulations, and therefore only Apixaban(S4) was subjected to FMO calculations. In Apixaban(S4), the phenyloxopiperidine scaffold adopts a conformation that occupies the S4 pocket (see Fig. S-4 left). A single structure was considered for Rivaroxaban [25]. In this structure, the chlorothiophene moiety occupies the S1 pocket, whereas the moiety containing the morpholinone and benzene rings occupies the S4 pocket (Fig. S-4 right). MD simulation MD simulations for each system were performed with GROMACS [74] in the NPT ensemble at 1 atm and 310 K. The Amber14SB Force Field [75] and the General Amber Force Field (GAFF2) [76] were used for the protein (factor Xa) and the ligands (four DOACs), respectively. Partial charges of the ligands were calculated at the Hartree–Fock (HF) level [77] with the 6‑31G(d) basis set [77–80] using the GAUSSIAN 16 program [81] and the restrained electrostatic potential (RESP) fitting scheme [82]. The TIP3P water model [83] was used for the solvent. Hydration and ionic neutralization were introduced by adding Na+ and Cl⁻ ions at a concentration of 0.15 M [67]. Periodic boundary conditions were imposed on the systems, and the temperature and pressure were maintained using the velocity‑rescaling thermostat and the Parrinello-Rahman barostat [84], respectively. The LINCS algorithm [85] was applied to constrain covalent bonds, with an integration time step of 2.0 fs. Long‑range Coulomb interactions were treated using the particle mesh Ewald method [86] with a real‑space cutoff of 12.0 Å, and van der Waals interactions were calculated with the same 12.0 Å cutoff. Each MD run was carried out for 200 ns. From the MD trajectories from 100 to 200 ns, a total of 101 droplet snapshots were extracted at 1 ns interval; although this sampling length is shorter than that employed in more extensive MD studies (e.g., Chen et al. [87, 88]), it is sufficient to generate representative structures for the subsequent FMO calculations [65–69]. For each snapshot, water molecules located more than 4 Å from the factor Xa-ligand complex were removed, while all ions were retained, following our previous MD-FMO protocols [63, 67]. Figure 2 illustrates representative snapshot structures (depicted by MOE [72]) for four DOAC ligands; Edoxaban and Betrixaban are of neutral type here. Important residues of the binding pockets, such as Trp215 and Gly218, as indicated in Ref. [37], are visible.​ These snapshot structures of the DOAC ligands can be cross-referenced in the corresponding subsections of the Results and Discussion section.Fig. 2Representative snapshot structures extracted from equilibrated MD simulations for A Edoxaban, B Betrixaban, C Apixaban, and D Rivaroxaban. The side chains of key residues (in gray) and the ligands (in green) are shown as stick models Representative snapshot structures extracted from equilibrated MD simulations for A Edoxaban, B Betrixaban, C Apixaban, and D Rivaroxaban. The side chains of key residues (in gray) and the ligands (in green) are shown as stick models FMO calculation FMO calculations were performed using the ABINIT-MP program [55, 89]. The calculations employed the FMO-MP2 level [90–92], which incorporates electron correlation at the second-order Møller-Plesset perturbation level [77]. The 6-31G(d) basis set was used as in our previous studies [63–70]. To compensate for the tendency of MP2 to overestimate interaction energies [93], a partial renormalization (PR) correction [94] was applied. A standard PIEDA scheme [57, 58] was adopted [89]; the electrostatic (ES), exchange-repulsion (EX), and charge-transfer (CT) components were evaluated at the HF level, whereas the dispersion interaction (DI) component was obtained from the PR-MP2 correlation energy.​ Dynamical averaging over the droplet models was performed for the IFIE and each PIEDA component. Each ligand-residue distance was evaluated as the distance between their geometric centers. Note that the FMO job completion rate for the samples ranged from 0.89 to 0.98, except for Edoxaban(N+), for which the rate was 0.54. Nevertheless, the averaging over the available snapshots is considered acceptable. In the normal FMO scheme [52–55], fragmentation is performed not at the peptide bond itself but between the sp3-hybridized α-carbon (with the side-chain) and the carbonyl carbon in the main-chain, using a pair of bond-detachment (BDA) and bond-attachment (BAA) atoms [95]. Consequently, care must be taken because, in cases where the main-chain carbonyl oxygen is involved in an interaction, the formally assigned interacting residue shifts toward the C-terminal side; see Ref. [96] for an example. In the interaction analyses in this study, a sharp symbol (#) is used to identify such right-shifted residues [97]. It should also be noted that the disulfide bond was kept intact according to the standard FMO fragmentation convention [52–55], and thus the interaction energy is given as the sum over the two cysteine residues; a trial attempt to treat these residues separately has recently been reported [98]. MOE [72] was used to visually analyze the interactions within the binding pockets. Results and discussion Edoxaban Figure 3 shows the average IFIE values for Edoxaban(N) by residue, respectively; an average IFIE ≤ − 3 kcal/mol is used as the listing threshold. The corresponding PIEDA results are provided in Fig. S-5 of SI, and the numerical values are summarized in Table S-1. The residue that contributed most to stabilization is Cys191#, which exhibits an average IFIE more than twice that of the other residues. However, as noted earlier [52–55], Cys191# was treated together with Cys220 because of the disulfide bond between them. Furthermore, the # symbol indicates that the interaction also includes the hydrogen-bonding contribution from the carbonyl oxygen of Gly219; refer to Fig. S-6 when necessary.Fig. 3Residues showing significant interactions with Edoxaban(N) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Residues showing significant interactions with Edoxaban(N) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Gly216 and Gln192 also show significant stabilizing contributions of − 8.9 and − 6.9 kcal/mol, respectively, but their standard deviations (approximately 5 kcal/mol) are large, indicating substantial thermal fluctuations. Trp215, Tyr99, and Phe174 exhibit considerable DI contributions (π-π and CH-π), as expected for residues with aromatic electrons; the importance of these residues was also denoted in Refs. [12] and [37]. Asp189 (in the S1 pocket) was addressed in Ref. [12], and this residue forms a long-range electrostatic interaction, with a distance of more than 6 Å from the ligand. Next, the results for Edoxaban(N+) are shown in Fig. 4. The corresponding PIEDA results are presented in Fig. S-7 and Table S-2. Comparison with Edoxaban(N) is shown in Fig. 5 as a scatter plot of mean IFIE values; note that a similar MD-based figure was reported in Ref. [37]. It is notable that Cys191# does not contribute to stabilization. In addition, the number of residues contributing to stabilization increases, with six residues—Glu97, Glu217#, Trp215, Thr98, Tyr99, and Phe174—being particularly prominent; recall [96, 97] that the hydrogen-bonding contribution from the carbonyl oxygen of Gly216 is included in Glu217#. All of these residues are located near S4; see Fig. S-8. In Edoxaban(N+), the electrostatic interaction energy contributions at Trp215, Tyr99, and Phe174 increased, which is attributed to cation-π interactions arising from the positive charge on the nitrogen atom. Thr98 contributes not only to the ES (− 16.1 kcal/mol) component but also to CT (− 1.8 kcal/mol) and DI (− 2.7 kcal/mol), indicating interactions with multiple components.Fig. 4Residues showing significant interactions with Edoxaban(N+) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviationFig. 5Scatter plot of mean IFIE values for residues comparing Edoxaban(N) and Edoxaban(N+). No filtering based on contribution was applied to this plot Residues showing significant interactions with Edoxaban(N+) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Scatter plot of mean IFIE values for residues comparing Edoxaban(N) and Edoxaban(N+). No filtering based on contribution was applied to this plot Betrixaban The results for Betrixaban(N) are shown in Fig. 6, and the corresponding PIEDA results are presented in Fig. S-9 and Table S-3. The residues contributing to stabilization were similar to those for Edoxaban(N), with Cys191# again making the largest contribution. However, characteristic changes were observed at Glu217#, Asp189, and Glu147. Glu217# exhibited an almost twofold increase in average IFIE, accompanied by a marked enhancement in ES energy. In addition, the contribution from CT energy (− 1.0 kcal/mol) suggests the formation of a hydrogen-bonding. Likewise, Asp189 also showed an increase in electrostatic interaction energy. Glu147 makes little contribution to the interaction with Edoxaban(N) but contributes substantially for Betrixaban(N), which can be attributed to its 1.3 Å shorter distance from the ligand.Fig. 6Residues showing significant interactions with Betrixaban(N) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Residues showing significant interactions with Betrixaban(N) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Results for Betrixaban(N+) are shown in Fig. 7, and PIEDA results are provided in Fig. S-10 and Table S-4. Glu217# exhibits the greatest stabilization, with electrostatic interactions dominant according to the PIEDA analysis. Similar to Edoxaban(N+), Glu97 also contributes significantly to stabilization, and its average IFIE is larger than that for Edoxaban(N+). Betrixaban possesses a charged dimethylamino group near the S4 porcket, suggesting enhanced stabilization through its interaction with Glu97 (see Fig. S-11). In contrast to Edoxaban(N+), Cys191# makes a meaningful contribution to stabilization, as it does for Betrixaban(N). Figure 8 shows the average IFIE values over residues for Betrixaban(N) and Betrixaban(N+), and the differences are of interest in comparison with Fig. 5 for Edoxaban. It is notable that the interactions involving Glu97 and Glu217# are substantially enhanced.Fig. 7Residues showing significant interactions with Betrixaban(N+) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviationFig. 8Scatter plot of mean IFIE values for residues comparing Betrixaban(N) and Betrixaban(N+). No filtering based on contribution was applied to this plot Residues showing significant interactions with Betrixaban(N+) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Scatter plot of mean IFIE values for residues comparing Betrixaban(N) and Betrixaban(N+). No filtering based on contribution was applied to this plot Apixaban Figure 9 and Fig. S-12 show the average IFIE and PIEDA results for Apixaban(S4) by residue, respectively, and the corresponding numerical values are summarized in Table S-5. The stabilizing residues are similar to those for the other neutral ligands [12]. Notably, the contribution from Cys191# is smaller than that for Edoxaban(N) and Betrixaban(N), reflecting its longer average distance from Apixaban(S4); see again Tables S-1 and S-3. It is noteworthy that Thr98 is contributive here, with ES and DI values of − 2.4 and − 2.2 kcal/mol, respectively.Fig. 9Residues showing significant interactions with Apixaban(S4) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Residues showing significant interactions with Apixaban(S4) (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Rivaroxaban Finally, Fig. 10 and Fig. S-13 show the average IFIE and PIEDA results for Rivaroxaban by residue, respectively, and the corresponding numerical values are listed in Table S-6. For this ligand as well, the residue contributing most to stabilization is Cys191# (associated with the S1 pocket), with substantial ES and DI contributions of − 21.5 and − 11.5 kcal/mol, respectively. Next, Tyr99 and Trp215 (both located in the S4 pocket) contribute to stabilization through CH-π interactions. Notably, Glu217# contributes less to the stabilization of rivaroxaban than to that of the other three neutral ligands. Since the average distance remains largely unchanged, this may reflect the size of the portion engaging the S1 pocket. (see Figs. S-3, S-4 and S-5 when needed).Fig. 10Residues showing significant interactions with Rivaroxaban (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Residues showing significant interactions with Rivaroxaban (average IFIE ≤ − 3 kcal/mol). Vertical bars indicate the standard deviation Interaction with Ser195 Ref. [34] focuses its analysis on Ser195, based on the presence of a serine protease structure in the active site of factor Xa [4] and the central catalytic role of Ser195 in the protease mechanism. It is therefore interesting to examine the interactions between Ser195 and each DOAC ligand. The role of water molecules around Ser195 was also addressed in Ref. [34]. Comparative analysis using droplet models of the uncharged ligands revealed that water molecules intervened at many time steps for Edoxaban(N) and Betrixaban(N). In contrast, no water intervention was observed for Rivaroxaban or Apixaban(S4). For ligands in which water intervention occurred, the nitrogen atom in the pyridine ring acted as an electron donor: refer to Fig. S-14. The IFIE results are plotted in Fig. 11, and the numerical values are summarized in Table S-7. For the IFIE values, Edoxaban(N) and Betrixaban(N) are larger than Apixaban(S4) and Rivaroxaban, primarily due to stabilization by the ES component. It is noteworthy that the effect of charge for Edoxaban(N+) and Betrixaban(N+) is not particularly large in this case.Fig. 11IFIE between Ser195 and each DOAC ligand. Vertical bars indicate the standard deviation IFIE between Ser195 and each DOAC ligand. Vertical bars indicate the standard deviation Leading residues and ligand‑specific interaction features The interactions between individual residues in the S1 and S4 pockets and the DOAC ligands were discussed previously (Refs. [12] and [37]). Our MD-FMO calculations provided comparable residue-specific analyses, offering dynamically averaged interaction energies and quantum-chemically resolved stabilization types identified by PIEDA components. Table 1 summarizes the characteristic interaction types between the four DOAC ligands (including two protonation states) and six key residues – Cys191#, Tyr99, Trp215, Glu217#, Phe174, and Gln192—that make significant contributions; these residues were also highlighted in Refs. [12] and [37]. Notably, the ES, CT, and DI components, reflecting hydrogen-bonding and π-electron-mediated attractive interactions [96], act cooperatively within the binding pockets [34, 37]. When appropriate, readers may also refer to Table S-8 for other contributive residues; Table 1 is a concise subset of the more exhaustive summary provided in Table S-8.Table 1Summary of PIEDA results for six key residues across the four DOAC ligands.Residue/PocketEdo.(N)Edo.(N+)Betri.(N)Betri.(N+)Api.(S4)RivaroTyr99/S4ES·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIPhe174/S4ES·CT·DIES·CT·DICT·DICT·DICT·DICys191#/S1ES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIGln192/S1ES·CT·DIES·CT·DICT·DICT·DICT·DIDITrp215/S4CT·DIES·CT·DICT·DIES·CT·DICT·DIES·CT·DIGlu217#/S4ES·DIES·DIES·DIES·DIES·DIDIA: Edo., Edoxaban; Betri., Betrixaban; Api., Apixaban; Rivaro., Rivaroxaban. Interaction labels for each residue are assigned on the basis of the individual PIEDA components Table 1Summary of PIEDA results for six key residues across the four DOAC ligands.Residue/PocketEdo.(N)Edo.(N+)Betri.(N)Betri.(N+)Api.(S4)RivaroTyr99/S4ES·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIPhe174/S4ES·CT·DIES·CT·DICT·DICT·DICT·DICys191#/S1ES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIES·CT·DIGln192/S1ES·CT·DIES·CT·DICT·DICT·DICT·DIDITrp215/S4CT·DIES·CT·DICT·DIES·CT·DICT·DIES·CT·DIGlu217#/S4ES·DIES·DIES·DIES·DIES·DIDIA: Edo., Edoxaban; Betri., Betrixaban; Api., Apixaban; Rivaro., Rivaroxaban. Interaction labels for each residue are assigned on the basis of the individual PIEDA components Summary of PIEDA results for six key residues across the four DOAC ligands. A: Edo., Edoxaban; Betri., Betrixaban; Api., Apixaban; Rivaro., Rivaroxaban. Interaction labels for each residue are assigned on the basis of the individual PIEDA components In addition to common interaction features, ligand‑dependent differences were also observed in the relative contributions of PIEDA components. For example, variations in the balance of ES and CT interactions versus DI contributions reflect differences in functional groups and aromatic moieties among the DOAC ligands, leading to distinct residue‑level interaction patterns within the S1 and S4 pockets. In particular, Edoxaban [22] exhibits relatively rich and diverse combinations of ES, CT and DI contributions among the key residues, suggesting a more versatile interaction profile within the S1 and S4 pockets. Conclusions The present study investigated complexes formed between factor Xa and four DOAC ligands (Edoxaban [22], Betrixaban [23], Apixaban [24], and Rivaroxaban [25]). Droplet-structure samples were generated from MD simulations to account for hydration and structural fluctuations. These samples were then subjected to standard FMO-MP2/6-31G(d) calculations, following an established MD-FMO workflow [63–70]. Finally, dynamically averaged interaction analyses of IFIE [56] and PIEDA [57, 58] were performed. Cys191#, Tyr99, Trp215, Glu217#, Phe174, and Gln192 were identified as major residues involved in stabilizing interactions across the four DOAC ligands. These residues were highlighted in previous studies (e.g., Refs. [12] and [37]), and their importance has now been reaffirmed here by our MD-FMO combination calculations. Consistent quantum–mechanical views of hydrogen-bonding and dispersion contributions were also obtained. As demonstrated in previous studies [63–70], FMO calculations enable a physically grounded decomposition of protein–ligand interactions—including electrostatic (ES), exchange repulsion (EX), charge-transfer (CT) and dispersion interaction (DI) components—beyond what is accessible by docking or conventional MD analyses alone. By incorporating multiple conformations derived from MD simulations, the present approach allows identification of key residues (or hot spots) governing ligand binding in a dynamic context, providing insights that are useful for rational drug design [71]. The FMODB database [99], developed and operated by the FMODD Consortium [100], serves as a repository for FMO calculation results. The results of the present calculations and analyses are also scheduled for registration. We conclude this paper with the hope that fundamental MD-FMO statistical interaction data will facilitate deeper mechanistic understanding and support the rational design of new and improved DOACs. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1