PyCoM: a python library for large-scale analysis of residue-residue coevolution data.

MOTIVATION: Computational methods to detect correlated amino acid positions in proteins have become a valuable tool to predict intra- and inter-residue protein contacts, protein structures, and effects of mutation on protein stability and function. While there are many tools and webservers to compute coevolution scoring matrices, there is no central repository of alignments and coevolution matrices for large-scale studies and pattern detection leveraging on biological and structural annotations already available in UniProt. RESULTS: We present a Python library, PyCoM, which enables users to query and analyze coevolution matrices and sequence alignments of 457 622 proteins, selected from UniProtKB/Swiss-Prot database (length ≤ 500 residues), from a precompiled coevolution matrix database (PyCoMdb). PyCoM facilitates the development of statistical analyses of residue coevolution patterns using filters on biological and structural annotations from UniProtKB/Swiss-Prot, with simple access to PyCoMdb for both novice and advanced users, supporting Jupyter Notebooks, Python scripts, and a web API access. The resource is open source and will help in generating data-driven computational models and methods to study and understand protein structures, stability, function, and design. AVAILABILITY AND IMPLEMENTATION: PyCoM code is freely available from https://github.com/scdantu/pycom and PyCoMdb and the Jupyter Notebook tutorials are freely available from https://pycom.brunel.ac.uk.

MDSubSampler: a posteriori sampling of important protein conformations from biomolecular simulations.

MOTIVATION: Molecular dynamics (MD) simulations have become routine tools for the study of protein dynamics and function. Thanks to faster GPU-based algorithms, atomistic and coarse-grained simulations are being used to explore biological functions over the microsecond timescale, yielding terabytes of data spanning multiple trajectories, thereby extracting relevant protein conformations without losing important information is often challenging. RESULTS: We present MDSubSampler, a Python library and toolkit for a posteriori subsampling of data from multiple trajectories. This toolkit provides access to uniform, random, stratified, weighted sampling, and bootstrapping sampling methods. Sampling can be performed under the constraint of preserving the original distribution of relevant geometrical properties. Possible applications include simulations post-processing, noise reduction, and structures selection for ensemble docking. AVAILABILITY AND IMPLEMENTATION: MDSubSampler is freely available at https://github.com/alepandini/MDSubSampler, along with guidance on installation and tutorials on how it can be used.

Dynamic Profiling of ß-Coronavirus 3CL Mpro Protease Ligand-Binding Sites.

ß-coronavirus (CoVs) alone has been responsible for three major global outbreaks in the 21st century. The current crisis has led to an urgent requirement to develop therapeutics. Even though a number of vaccines are available, alternative strategies targeting essential viral components are required as a backup against the emergence of lethal viral variants. One such target is the main protease (Mpro) that plays an indispensable role in viral replication. The availability of over 270 Mpro X-ray structures in complex with inhibitors provides unique insights into ligand-protein interactions. Herein, we provide a comprehensive comparison of all nonredundant ligand-binding sites available for SARS-CoV2, SARS-CoV, and MERS-CoV Mpro. Extensive adaptive sampling has been used to investigate structural conservation of ligand-binding sites using Markov state models (MSMs) and compare conformational dynamics employing convolutional variational auto-encoder-based deep learning. Our results indicate that not all ligand-binding sites are dynamically conserved despite high sequence and structural conservation across ß-CoV homologs. This highlights the complexity in targeting all three Mpro enzymes with a single pan inhibitor.

Structure-based enzyme engineering improves donor-substrate recognition of Arabidopsis thaliana glycosyltransferases.

Glycosylation of secondary metabolites involves plant UDP-dependent glycosyltransferases (UGTs). UGTs have shown promise as catalysts in the synthesis of glycosides for medical treatment. However, limited understanding at the molecular level due to insufficient biochemical and structural information has hindered potential applications of most of these UGTs. In the absence of experimental crystal structures, we employed advanced molecular modeling and simulations in conjunction with biochemical characterization to design a workflow to study five Group H Arabidopsis thaliana (76E1, 76E2, 76E4, 76E5, 76D1) UGTs. Based on our rational structural manipulation and analysis, we identified key amino acids (P129 in 76D1; D374 in 76E2; K275 in 76E4), which when mutated improved donor substrate recognition than wildtype UGTs. Molecular dynamics simulations and deep learning analysis identified structural differences, which drive substrate preferences. The design of these UGTs with broader substrate specificity may play important role in biotechnological and industrial applications. These findings can also serve as basis to study other plant UGTs and thereby advancing UGT enzyme engineering.

Allosteric Priming of E. coli CheY by the Flagellar Motor Protein FliM.

Phosphorylation of Escherichia coli CheY protein transduces chemoreceptor stimulation to a highly cooperative flagellar motor response. CheY binds to the N-terminal peptide of the FliM motor protein (FliMN). Constitutively active D13K-Y106W CheY has been an important tool for motor physiology. The crystal structures of CheY and CheY ⋅ FliMN with and without D13K-Y106W have shown FliMN-bound CheY contains features of both active and inactive states. We used molecular dynamics (MD) simulations to characterize the CheY conformational landscape accessed by FliMN and D13K-Y106W. Mutual information measures identified the central features of the long-range CheY allosteric network between D57 phosphorylation site and the FliMN interface, namely the closure of the α4-ß4 hinge and inward rotation of Y- or W106 with W58. We used hydroxy-radical foot printing with mass spectroscopy (XFMS) to track the solvent accessibility of these and other side chains. The solution XFMS oxidation rate correlated with the solvent-accessible area of the crystal structures. The protection of allosteric relay side chains reported by XFMS confirmed the intermediate conformation of the native CheY ⋅ FliMN complex, the inactive state of free D13K-Y106W CheY, and the MD-based network architecture. We extended the MD analysis to determine temporal coupling and energetics during activation. Coupled aromatic residue rotation was a graded rather than a binary switch, with Y- or W106 side-chain burial correlated with increased FliMN affinity. Activation entrained CheY fold stabilization to FliMN affinity. The CheY network could be partitioned into four dynamically coordinated sectors. Residue substitutions mapped to sectors around D57 or the FliMN interface according to phenotype. FliMN increased sector size and interactions. These sectors fused between the substituted K13-W106 residues to organize a tightly packed core and novel surfaces that may bind additional sites to explain the cooperative motor response. The community maps provide a more complete description of CheY priming than proposed thus far.

Conformational coupling by trans-phosphorylation in calcium calmodulin dependent kinase II.

The calcium calmodulin-dependent protein kinase II (CaMKII) is a dodecameric holoenzyme important for encoding memory. Its activation, triggered by binding of calcium-calmodulin, persists autonomously after calmodulin dissociation. One (receiver) kinase captures and subsequently phosphorylates the regulatory domain peptide of a donor kinase forming a chained dimer as the first stage of autonomous activation. Protein dynamics simulations examined the conformational changes triggered by dimer formation and phosphorylation, aimed to provide a molecular rationale for human mutations that result in learning disabilities. Ensembles generated from X-ray crystal structures were characterized by network centrality and community analysis. Mutual information related collective motions to local fragment dynamics encoded with a structural alphabet. Implicit solvent tCONCOORD conformational ensembles revealed the dynamic architecture of inactive kinase domains was co-opted in the activated dimer but the network hub shifted from the nucleotide binding cleft to the captured peptide. Explicit solvent molecular dynamics (MD) showed nucleotide and substrate binding determinants formed coupled nodes in long-range signal relays between regulatory peptides in the dimer. Strain in the extended captured peptide was balanced by reduced flexibility of the receiver kinase C-lobe core. The relays were organized around a hydrophobic patch between the captured peptide and a key binding helix. The human mutations aligned along the relays. Thus, these mutations could disrupt the allosteric network alternatively, or in addition, to altered binding affinities. Non-binding protein sectors distant from the binding sites mediated the allosteric signalling; providing possible targets for inhibitor design. Phosphorylation of the peptide modulated the dielectric of its binding pocket to strengthen the patch, non-binding sectors, domain interface and temporal correlations between parallel relays. These results provide the molecular details underlying the reported positive kinase cooperativity to enrich the discussion on how autonomous activation by phosphorylation leads to long-term behavioural effects.

In silico identification of rescue sites by double force scanning.

Motivation: A deleterious amino acid change in a protein can be compensated by a second-site rescue mutation. These compensatory mechanisms can be mimicked by drugs. In particular, the location of rescue mutations can be used to identify protein regions that can be targeted by small molecules to reactivate a damaged mutant. Results: We present the first general computational method to detect rescue sites. By mimicking the effect of mutations through the application of forces, the double force scanning (DFS) method identifies the second-site residues that make the protein structure most resilient to the effect of pathogenic mutations. We tested DFS predictions against two datasets containing experimentally validated and putative evolutionary-related rescue sites. A remarkably good agreement was found between predictions and experimental data. Indeed, almost half of the rescue sites in p53 was correctly predicted by DFS, with 65% of remaining sites in contact with DFS predictions. Similar results were found for other proteins in the evolutionary dataset. Availability and implementation: The DFS code is available under GPL at https://fornililab.github.io/dfs/. Supplementary information: Supplementary data are available at Bioinformatics online.

Ligand-induced perturbation of the HIF-2α:ARNT dimer dynamics.

Hypoxia inducible factors (HIFs) are transcription factors belonging to the basic helix-loop-helix PER-ARNT-SIM (bHLH-PAS) protein family with a role in sensing oxygen levels in the cell. Under hypoxia, the HIF-α degradation pathway is blocked and dimerization with the aryl hydrocarbon receptor nuclear translocator (ARNT) makes HIF-α transcriptionally active. Due to the common hypoxic environment of tumors, inhibition of this mechanism by destabilization of HIF-α:ARNT dimerization has been proposed as a promising therapeutic strategy. Following the discovery of a druggable cavity within the PAS-B domain of HIF-2α, research efforts have been directed to identify artificial ligands that can impair heterodimerization. Although the crystallographic structures of the HIF-2α:ARNT complex have elucidated the dimer architecture and the 0X3-inhibitor placement within the HIF-2α PAS-B, unveiling the inhibition mechanism requires investigation of how ligand-induced perturbations could dynamically propagate through the structure and affect dimerization. To this end, we compared evolutionary features, intrinsic dynamics and energetic properties of the dimerization interfaces of HIF-2α:ARNT in both the apo and holo forms. Residue conservation analysis highlighted inter-domain connecting elements that have a role in dimerization. Analysis of domain contributions to the dimerization energy demonstrated the importance of bHLH and PAS-A of both partners and of HIF-2α PAS-B domain in dimer stabilization. Among quaternary structure oscillations revealed by Molecular Dynamics simulations, the hinge-bending motion of the ARNT PAS-B domain around the flexible PAS-A/PAS-B linker supports a general model for ARNT dimerization in different heterodimers. Comparison of the HIF-2α:ARNT dynamics in the apo and 0X3-bound forms indicated a model of inhibition where the HIF-2α-PAS-B interfaces are destabilised as a result of water-bridged ligand-protein interactions and these local effects allosterically propagate to perturb the correlated motions of the domains and inter-domain communication. These findings will guide the design of improved inhibitors to contrast cell survival in tumor masses.

The Phylogenetic Signature Underlying ATP Synthase c-Ring Compliance.

The proton-driven ATP synthase (FOF1) is comprised of two rotary, stepping motors (FO and F1) coupled by an elastic power transmission. The elastic compliance resides in the rotor module that includes the membrane-embedded FO c-ring. Proton transport by FO is firmly coupled to the rotation of the c-ring relative to other FO subunits (ab2). It drives ATP synthesis. We used a computational method to investigate the contribution of the c-ring to the total elastic compliance. We performed principal component analysis of conformational ensembles built using distance constraints from the bovine mitochondrial c-ring x-ray structure. Angular rotary twist, the dominant ring motion, was estimated to show that the c-ring accounted in part for the measured compliance. Ring rotation was entrained to rotation of the external helix within each hairpin-shaped c-subunit in the ring. Ensembles of monomer and dimers extracted from complete c-rings showed that the coupling between collective ring and the individual subunit motions was independent of the size of the c-ring, which varies between organisms. Molecular determinants were identified by covariance analysis of residue coevolution and structural-alphabet-based local dynamics correlations. The residue coevolution gave a readout of subunit architecture. The dynamic couplings revealed that the hinge for both ring and subunit helix rotations was constructed from the proton-binding site and the adjacent glycine motif (IB-GGGG) in the midmembrane plane. IB-GGGG motifs were linked by long-range couplings across the ring, while intrasubunit couplings connected the motif to the conserved cytoplasmic loop and adjacent segments. The correlation with principal collective motions shows that the couplings underlie both ring rotary and bending motions. Noncontact couplings between IB-GGGG motifs matched the coevolution signal as well as contact couplings. The residue coevolution reflects the physiological importance of the dynamics that may link proton transfer to ring compliance.

BCR-ABL residues interacting with ponatinib are critical to preserve the tumorigenic potential of the oncoprotein.

Patients with chronic myeloid leukemia in whom tyrosine kinase inhibitors (TKIs) fail often present mutations in the BCR-ABL catalytic domain. We noticed a lack of substitutions involving 4 amino acids (E286, M318, I360, and D381) that form hydrogen bonds with ponatinib. We therefore introduced mutations in each of these residues, either preserving or altering their physicochemical properties. We found that E286, M318, I360, and D381 are dispensable for ABL and BCR-ABL protein stability but are critical for preserving catalytic activity. Indeed, only a "conservative" I360T substitution retained kinase proficiency and transforming potential. Molecular dynamics simulations of BCR-ABL(I360T) revealed differences in both helix αC dynamics and protein-correlated motions, consistent with a modified ATP-binding pocket. Nevertheless, this mutant remained sensitive to ponatinib, imatinib, and dasatinib. These results suggest that changes in the 4 BCR-ABL residues described here would be selected against by a lack of kinase activity or by maintained responsiveness to TKIs. Notably, amino acids equivalent to those identified in BCR-ABL are conserved in 51% of human tyrosine kinases. Hence, these residues may represent an appealing target for the design of pharmacological compounds that would inhibit additional oncogenic tyrosine kinases while avoiding the emergence of resistance due to point mutations.

GSATools: analysis of allosteric communication and functional local motions using a structural alphabet.

MOTIVATION: GSATools is a free software package to analyze conformational ensembles and to detect functional motions in proteins by means of a structural alphabet. The software integrates with the widely used GROMACS simulation package and can generate a range of graphical outputs. Three applications can be supported: (i) investigation of the conformational variability of local structures; (ii) detection of allosteric communication; and (iii) identification of local regions that are critical for global functional motions. These analyses provide insights into the dynamics of proteins and allow for targeted design of functional mutants in theoretical and experimental studies. AVAILABILITY: The C source code of the GSATools, along with a set of pre-compiled binaries, is freely available under GNU General Public License from http://mathbio.nimr.mrc.ac.uk/wiki/GSATools.

N4BP1 functions as a dimerization-dependent linear ubiquitin reader which regulates TNF signalling.

Signalling through TNFR1 modulates proinflammatory gene transcription and programmed cell death, and its impairment causes autoimmune diseases and cancer. NEDD4-binding protein 1 (N4BP1) is a critical suppressor of proinflammatory cytokine production that acts as a regulator of innate immune signalling and inflammation. However, our current understanding about the molecular properties that enable N4BP1 to exert its suppressive potential remain limited. Here, we show that N4BP1 is a novel linear ubiquitin reader that negatively regulates NFκB signalling by its unique dimerization-dependent ubiquitin-binding module that we named LUBIN. Dimeric N4BP1 strategically positions two non-selective ubiquitin-binding domains to ensure preferential recognition of linear ubiquitin. Under proinflammatory conditions, N4BP1 is recruited to the nascent TNFR1 signalling complex, where it regulates duration of proinflammatory signalling in LUBIN-dependent manner. N4BP1 deficiency accelerates TNFα-induced cell death by increasing complex II assembly. Under proapoptotic conditions, caspase-8 mediates proteolytic processing of N4BP1, resulting in rapid degradation of N4BP1 by the 26 S proteasome, and acceleration of apoptosis. In summary, our findings demonstrate that N4BP1 dimerization creates a novel type of ubiquitin reader that selectively recognises linear ubiquitin which enables the timely and coordinated regulation of TNFR1-mediated inflammation and cell death.

Comparative analysis of homology models of the AH receptor ligand binding domain: verification of structure-function predictions by site-directed mutagenesis of a nonfunctional receptor.

The aryl hydrocarbon receptor (AHR) is a ligand-dependent transcription factor that mediates the biological and toxic effects of a wide variety of structurally diverse chemicals, including the toxic environmental contaminant 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). While significant interspecies differences in AHR ligand binding specificity, selectivity, and response have been observed, the structural determinants responsible for those differences have not been determined, and homology models of the AHR ligand-binding domain (LBD) are available for only a few species. Here we describe the development and comparative analysis of homology models of the LBD of 16 AHRs from 12 mammalian and nonmammalian species and identify the specific residues contained within their ligand binding cavities. The ligand-binding cavity of the fish AHR exhibits differences from those of mammalian and avian AHRs, suggesting a slightly different TCDD binding mode. Comparison of the internal cavity in the LBD model of zebrafish (zf) AHR2, which binds TCDD with high affinity, to that of zfAHR1a, which does not bind TCDD, revealed that the latter has a dramatically shortened binding cavity due to the side chains of three residues (Tyr296, Thr386, and His388) that reduce the amount of internal space available to TCDD. Mutagenesis of two of these residues in zfAHR1a to those present in zfAHR2 (Y296H and T386A) restored the ability of zfAHR1a to bind TCDD and to exhibit TCDD-dependent binding to DNA. These results demonstrate the importance of these two amino acids and highlight the predictive potential of comparative analysis of homology models from diverse species. The availability of these AHR LBD homology models will facilitate in-depth comparative studies of AHR ligand binding and ligand-dependent AHR activation and provide a novel avenue for examining species-specific differences in AHR responsiveness.

Detection of allosteric signal transmission by information-theoretic analysis of protein dynamics.

Allostery offers a highly specific way to modulate protein function. Therefore, understanding this mechanism is of increasing interest for protein science and drug discovery. However, allosteric signal transmission is difficult to detect experimentally and to model because it is often mediated by local structural changes propagating along multiple pathways. To address this, we developed a method to identify communication pathways by an information-theoretical analysis of molecular dynamics simulations. Signal propagation was described as information exchange through a network of correlated local motions, modeled as transitions between canonical states of protein fragments. The method was used to describe allostery in two-component regulatory systems. In particular, the transmission from the allosteric site to the signaling surface of the receiver domain NtrC was shown to be mediated by a layer of hub residues. The location of hubs preferentially connected to the allosteric site was found in close agreement with key residues experimentally identified as involved in the signal transmission. The comparison with the networks of the homologues CheY and FixJ highlighted similarities in their dynamics. In particular, we showed that a preorganized network of fragment connections between the allosteric and functional sites exists already in the inactive state of all three proteins.

PathDetect-SOM: A Neural Network Approach for the Identification of Pathways in Ligand Binding Simulations.

Understanding the process of ligand-protein recognition is important to unveil biological mechanisms and to guide drug discovery and design. Enhanced-sampling molecular dynamics is now routinely used to simulate the ligand binding process, resulting in the need for suitable tools for the analysis of large data sets of binding events. Here, we designed, implemented, and tested PathDetect-SOM, a tool based on self-organizing maps to build concise visual models of the ligand binding pathways sampled along single simulations or replicas. The tool performs a geometric clustering of the trajectories and traces the pathways over an easily interpretable 2D map and, using an approximate transition matrix, it can build a graph model of concurrent pathways. The tool was tested on three study cases representing different types of problems and simulation techniques. A clear reconstruction of the sampled pathways was derived in all cases, and useful information on the energetic features of the processes was recovered. The tool is available at https://github.com/MottaStefano/PathDetect-SOM.

Recognizing the Binding Pattern and Dissociation Pathways of the p300 Taz2-p53 TAD2 Complex.

The dynamic association and dissociation between proteins are the basis of cellular signal transduction. This process becomes much more complicated if one or both interaction partners are intrinsically disordered because intrinsically disordered proteins can undergo disorder-to-order transitions upon binding to their partners. p53, a transcription factor with disordered regions, plays significant roles in many cellular signaling pathways. It is critical to understand the binding/unbinding mechanism involving these disordered regions of p53 at the residue level to reveal how p53 performs its biological functions. Here, we studied the dissociation process of the intrinsically disordered N-terminal transactivation domain 2 (TAD2) of p53 and the transcriptional adaptor zinc-binding 2 (Taz2) domain of transcriptional coactivator p300 using a combination of classical molecular dynamics, steered molecular dynamics, self-organizing maps, and time-resolved force distribution analysis (TRFDA). We observed two different dissociation pathways with different probabilities. One dissociation pathway starts from the TAD2 N-terminus and propagates to the α-helix and finally the C-terminus. The other dissociation pathway is in the opposite order. Subsequent TRFDA results reveal that key residues in TAD2 play critical roles. Besides the residues in agreement with previous experimental results, we also highlighted some other residues that play important roles in the disassociation process. In the dissociation process, non-native interactions were formed to partially compensate for the energy loss due to the breaking of surrounding native interactions. Moreover, our statistical analysis results of other experimentally determined complex structures involving either Taz2 or TAD2 suggest that the binding of the Taz2-TAD2 complex is mainly governed by the binding site of Taz2, which includes three main binding regions. Therefore, the complexes involving Taz2 may follow similar binding/unbinding behaviors, which could be studied together to generate common principles.

Conformational and functional analysis of molecular dynamics trajectories by self-organising maps.

BACKGROUND: Molecular dynamics (MD) simulations are powerful tools to investigate the conformational dynamics of proteins that is often a critical element of their function. Identification of functionally relevant conformations is generally done clustering the large ensemble of structures that are generated. Recently, Self-Organising Maps (SOMs) were reported performing more accurately and providing more consistent results than traditional clustering algorithms in various data mining problems. We present a novel strategy to analyse and compare conformational ensembles of protein domains using a two-level approach that combines SOMs and hierarchical clustering. RESULTS: The conformational dynamics of the α-spectrin SH3 protein domain and six single mutants were analysed by MD simulations. The Cα's Cartesian coordinates of conformations sampled in the essential space were used as input data vectors for SOM training, then complete linkage clustering was performed on the SOM prototype vectors. A specific protocol to optimize a SOM for structural ensembles was proposed: the optimal SOM was selected by means of a Taguchi experimental design plan applied to different data sets, and the optimal sampling rate of the MD trajectory was selected. The proposed two-level approach was applied to single trajectories of the SH3 domain independently as well as to groups of them at the same time. The results demonstrated the potential of this approach in the analysis of large ensembles of molecular structures: the possibility of producing a topological mapping of the conformational space in a simple 2D visualisation, as well as of effectively highlighting differences in the conformational dynamics directly related to biological functions. CONCLUSIONS: The use of a two-level approach combining SOMs and hierarchical clustering for conformational analysis of structural ensembles of proteins was proposed. It can easily be extended to other study cases and to conformational ensembles from other sources.

Predicting the accuracy of protein-ligand docking on homology models.

Ligand-protein docking is increasingly used in Drug Discovery. The initial limitations imposed by a reduced availability of target protein structures have been overcome by the use of theoretical models, especially those derived by homology modeling techniques. While this greatly extended the use of docking simulations, it also introduced the need for general and robust criteria to estimate the reliability of docking results given the model quality. To this end, a large-scale experiment was performed on a diverse set including experimental structures and homology models for a group of representative ligand-protein complexes. A wide spectrum of model quality was sampled using templates at different evolutionary distances and different strategies for target-template alignment and modeling. The obtained models were scored by a selection of the most used model quality indices. The binding geometries were generated using AutoDock, one of the most common docking programs. An important result of this study is that indeed quantitative and robust correlations exist between the accuracy of docking results and the model quality, especially in the binding site. Moreover, state-of-the-art indices for model quality assessment are already an effective tool for an a priori prediction of the accuracy of docking experiments in the context of groups of proteins with conserved structural characteristics.

Machine Learning Prediction of Allosteric Drug Activity from Molecular Dynamics.

Allosteric drugs have been attracting increasing interest over the past few years. In this context, it is common practice to use high-throughput screening for the discovery of non-natural allosteric drugs. While the discovery stage is supported by a growing amount of biological information and increasing computing power, major challenges still remain in selecting allosteric ligands and predicting their effect on the target protein's function. Indeed, allosteric compounds can act both as inhibitors and activators of biological responses. Computational approaches to the problem have focused on variations on the theme of molecular docking coupled to molecular dynamics with the aim of recovering information on the (long-range) modulation typical of allosteric proteins.

Reconstruction of ARNT PAS-B Unfolding Pathways by Steered Molecular Dynamics and Artificial Neural Networks.

Several experimental studies indicated that large conformational changes, including partial domain unfolding, have a role in the functional mechanisms of the basic helix loop helix Per/ARNT/SIM (bHLH-PAS) transcription factors. Recently, single-molecule atomic force microscopy (AFM) revealed two distinct pathways for the mechanical unfolding of the ARNT PAS-B. In this work we used steered molecular dynamics simulations to gain new insights into this process at an atomistic level. To reconstruct and classify pathways sampled in multiple simulations, we designed an original approach based on the use of self-organizing maps (SOMs). This led us to identify two types of unfolding pathways for the ARNT PAS-B, which are in good agreement with the AFM findings. Analysis of average forces mapped on the SOM revealed a stable conformation of the PAS-B along one pathway, which represents a possible structural model for the intermediate state detected by AFM. The approach here proposed will facilitate the study of other signal transmission mechanisms involving the folding/unfolding of PAS domains.