Developing a GNN-based AI model to predict mitochondrial toxicity using the bagging method.

Mitochondrial toxicity has been implicated in the development of various toxicities, including hepatotoxicity. Therefore, mitochondrial toxicity has become a major screening factor in the early discovery phase of drug development. Several models have been developed to predict mitochondrial toxicity based on chemical structures. However, they only provide a binary classification of positive or negative results and do not provide the substructures that contribute to a positive decision. Therefore, we developed an artificial intelligence (AI) model to predict mitochondrial toxicity and visualize structural alerts. To construct the model, we used the open-source software library kMoL, which employs a graph neural network approach that allows learning from chemical structure data. We also utilized the integrated gradient method, which enables the visualization of substructures that contribute to positive results. The dataset used to construct the AI model exhibited a significant imbalance, with significantly more negative than positive data. To address this, we employed the bagging method, which resulted in a model with high predictive performance, as evidenced by an F1 score of 0.839. This model can also be used to visualize substructures that contribute to mitochondrial toxicity using the integrated gradient method. Our AI model predicts mitochondrial toxicity based on chemical structures and may contribute to screening mitochondrial toxicity in the early stages of drug discovery.

A macrocyclic kinase inhibitor overcomes triple resistant mutations in EGFR-positive lung cancer.

Brigatinib-based therapy was effective against osimertinib-resistant EGFR C797S mutants and is undergoing clinical studies. However, tumor relapse suggests additional resistance mutations might emerge. Here, we first demonstrated the binding mode of brigatinib to the EGFR-T790M/C797S mutant by crystal structure analysis and predicted brigatinib-resistant mutations through a cell-based assay including N-ethyl-N-nitrosourea (ENU) mutagenesis. We found that clinically reported L718 and G796 compound mutations appeared, consistent with their proximity to the binding site of brigatinib, and brigatinib-resistant quadruple mutants such as EGFR-activating mutation/T790M/C797S/L718M were resistant to all the clinically available EGFR-TKIs. BI-4020, a fourth-generation EGFR inhibitor with a macrocyclic structure, overcomes the quadruple and major EGFR-activating mutants but not the minor mutants, such as L747P or S768I. Molecular dynamics simulation revealed the binding mode and affinity between BI-4020 and EGFR mutants. This study identified potential therapeutic strategies using the new-generation macrocyclic EGFR inhibitor to overcome the emerging ultimate resistance mutants.

VGAE-MCTS: A New Molecular Generative Model Combining the Variational Graph Auto-Encoder and Monte Carlo Tree Search.

Molecular generation is crucial for advancing drug discovery, materials science, and chemical exploration. It expedites the search for new drug candidates, facilitates tailored material creation, and enhances our understanding of molecular diversity. By employing artificial intelligence techniques such as molecular generative models based on molecular graphs, researchers have tackled the challenge of identifying efficient molecules with desired properties. Here, we propose a new molecular generative model combining a graph-based deep neural network and a reinforcement learning technique. We evaluated the validity, novelty, and optimized physicochemical properties of the generated molecules. Importantly, the model explored uncharted regions of chemical space, allowing for the efficient discovery and design of new molecules. This innovative approach has considerable potential to revolutionize drug discovery, materials science, and chemical research for accelerating scientific innovation. By leveraging advanced techniques and exploring previously unexplored chemical spaces, this study offers promising prospects for the efficient discovery and design of new molecules in the field of drug development.

Improving Compound-Protein Interaction Prediction by Self-Training with Augmenting Negative Samples.

Identifying compound-protein interactions (CPIs) is crucial for drug discovery. Since experimentally validating CPIs is often time-consuming and costly, computational approaches are expected to facilitate the process. Rapid growths of available CPI databases have accelerated the development of many machine-learning methods for CPI predictions. However, their performance, particularly their generalizability against external data, often suffers from a data imbalance attributed to the lack of experimentally validated inactive (negative) samples. In this study, we developed a self-training method for augmenting both credible and informative negative samples to improve the performance of models impaired by data imbalances. The constructed model demonstrated higher performance than those constructed with other conventional methods for solving data imbalances, and the improvement was prominent for external datasets. Moreover, examination of the prediction score thresholds for pseudo-labeling during self-training revealed that augmenting the samples with ambiguous prediction scores is beneficial for constructing a model with high generalizability. The present study provides guidelines for improving CPI predictions on real-world data, thus facilitating drug discovery.

Quantitative analysis of protein dynamics using a deep learning technique combined with experimental cryo-EM density data and MD simulations.

Protein functions associated with biological activity are precisely regulated by both tertiary structure and dynamic behavior. Thus, elucidating the high-resolution structures and quantitative information on in-solution dynamics is essential for understanding the molecular mechanisms. The main experimental approaches for determining tertiary structures include nuclear magnetic resonance (NMR), X-ray crystallography, and cryogenic electron microscopy (cryo-EM). Among these procedures, recent remarkable advances in the hardware and analytical techniques of cryo-EM have increasingly determined novel atomic structures of macromolecules, especially those with large molecular weights and complex assemblies. In addition to these experimental approaches, deep learning techniques, such as AlphaFold 2, accurately predict structures from amino acid sequences, accelerating structural biology research. Meanwhile, the quantitative analyses of the protein dynamics are conducted using experimental approaches, such as NMR and hydrogen-deuterium mass spectrometry, and computational approaches, such as molecular dynamics (MD) simulations. Although these procedures can quantitatively explore dynamic behavior at high resolution, the fundamental difficulties, such as signal crowding and high computational cost, greatly hinder their application to large and complex biological macromolecules. In recent years, machine learning techniques, especially deep learning techniques, have been actively applied to structural data to identify features that are difficult for humans to recognize from big data. Here, we review our approach to accurately estimate dynamic properties associated with local fluctuations from three-dimensional cryo-EM density data using a deep learning technique combined with MD simulations.

Novel Calcium-Binding Ablating Mutations Induce Constitutive RET Activity and Drive Tumorigenesis.

Distinguishing oncogenic mutations from variants of unknown significance (VUS) is critical for precision cancer medicine. Here, computational modeling of 71,756 RET variants for positive selection together with functional assays of 110 representative variants identified a three-dimensional cluster of VUSs carried by multiple human cancers that cause amino acid substitutions in the calmodulin-like motif (CaLM) of RET. Molecular dynamics simulations indicated that CaLM mutations decrease interactions between Ca2+ and its surrounding residues and induce conformational distortion of the RET cysteine-rich domain containing the CaLM. RET-CaLM mutations caused ligand-independent constitutive activation of RET kinase by homodimerization mediated by illegitimate disulfide bond formation. RET-CaLM mutants possessed oncogenic and tumorigenic activities that could be suppressed by tyrosine kinase inhibitors targeting RET. This study identifies calcium-binding ablating mutations as a novel type of oncogenic mutation of RET and indicates that in silico-driven annotation of VUSs of druggable oncogenes is a promising strategy to identify targetable driver mutations. SIGNIFICANCE: Comprehensive proteogenomic and in silico analyses of a vast number of VUSs identify a novel set of oncogenic and druggable mutations in the well-characterized RET oncogene.

Enhanced Conformational Sampling with an Adaptive Coarse-Grained Elastic Network Model Using Short-Time All-Atom Molecular Dynamics.

Compared to all-atom molecular dynamics (AA-MD) simulations, coarse-grained (CG) MD simulations can significantly reduce calculation costs. However, existing CG-MD methods are unsuitable for sampling structures that depart significantly from the initial structure without any biased force. In this study, we developed a new adaptive CG elastic network model (ENM), in which the dynamic cross-correlation coefficient based on short-time AA-MD of at most ns order is considered. By applying Bayesian optimization to search for a suitable parameter among the vast parameter space of adaptive CG-ENM, we succeeded in reducing the searching cost to approximately 10% of those for random sampling and exhaustive sampling. To evaluate the performance of adaptive CG-ENM, we applied the new methodology to adenylate kinase (ADK) and glutamine binding protein (GBP) in the apo state. The results showed that the structural ensembles explored by adaptive CG-ENM could be considerably more diverse than those by conventional ENMs with enhanced sampling such as temperature replica exchange MD and long-time AA-MD of 1 µs. In particular, some of the structures sampled by adaptive ENM are relatively close to the holo-type structures of ADK and GBP. Furthermore, as a challenging task, to demonstrate the advantages of the CG model with lower calculation cost, we applied our new methodology to a larger biomolecule, integrin (αV) in the inactive state. Then, we sampled various structural ensembles, including extended structures that are apparently different from inactive ones.

Integrated approach to functional analysis of an ERBB2 variant of unknown significance detected by a cancer gene panel test.

PURPOSE: Dealing with variants of unknown significance (VUS) is an important issue in the clinical application of NGS-based cancer gene panel tests. We detected a novel ERBB2 extracellular domain VUS, c.1157A > G p.(E401G), in a cancer gene panel test. Since the mechanisms of activation by ERBB2 extracellular domain (ECD) variants are not fully understood, we aimed to clarify those mechanisms and the biological functions of ERBB2 E401G. METHODS: ERBB2 E401G was selected as VUS for analysis because multiple software tools predicted its pathogenicity. We prepared ERBB2 expression vectors with the E401G variant as well as vectors with S310F and E321G, which are known to be activating mutations. On the basis of wild-type ERBB2 or mutant ERBB2 expression in cell lines without ERBB2 amplification or variants, we evaluated the phosphorylation of human epidermal growth factor receptor 2 and related proteins, and investigated with molecular dynamics (MD) simulation the mechanisms conferred by the variants. The biological effects of ERBB2 E401G were also investigated, both in vitro and in vivo. RESULTS: We found that ERBB2 E401G enhances C-terminal phosphorylation in a way similar to S310F. MD simulation analysis revealed that these variants maintain the stability of the EGFR-HER2 heterodimer in a ligand-independent manner. Moreover, ERBB2 E401G-transduced cells showed an increased invasive capacity in vitro and an increased tumor growth capacity in vivo. CONCLUSION: Our results provide important information on the activating mechanisms of ERBB2 extracellular domain (ECD) variants and illustrate a model workflow integrating wet and dry bench processes for the analysis of VUS detected with cancer gene panel tests.

Machine learning to estimate the local quality of protein crystal structures.

Low-resolution electron density maps can pose a major obstacle in the determination and use of protein structures. Herein, we describe a novel method, called quality assessment based on an electron density map (QAEmap), which evaluates local protein structures determined by X-ray crystallography and could be applied to correct structural errors using low-resolution maps. QAEmap uses a three-dimensional deep convolutional neural network with electron density maps and their corresponding coordinates as input and predicts the correlation between the local structure and putative high-resolution experimental electron density map. This correlation could be used as a metric to modify the structure. Further, we propose that this method may be applied to evaluate ligand binding, which can be difficult to determine at low resolution.

Structure-Based de Novo Molecular Generator Combined with Artificial Intelligence and Docking Simulations.

Recently, molecular generation models based on deep learning have attracted significant attention in drug discovery. However, most existing molecular generation models have serious limitations in the context of drug design wherein they do not sufficiently consider the effect of the three-dimensional (3D) structure of the target protein in the generation process. In this study, we developed a new deep learning-based molecular generator, SBMolGen, that integrates a recurrent neural network, a Monte Carlo tree search, and docking simulations. The results of an evaluation using four target proteins (two kinases and two G protein-coupled receptors) showed that the generated molecules had a better binding affinity score (docking score) than the known active compounds, and the generated molecules possessed a broader chemical space distribution. SBMolGen not only generates novel binding active molecules but also presents 3D docking poses with target proteins, which will be useful in subsequent drug design. The code is available at https://github.com/clinfo/SBMolGen.

Oncogenic mutations Q61L and Q61H confer active form-like structural features to the inactive state (state 1) conformation of H-Ras protein.

GTP-bound forms of Ras proteins (Ras•GTP) assume two interconverting conformations, "inactive" state 1 and "active" state 2. Our previous study on the crystal structure of the state 1 conformation of H-Ras in complex with guanosine 5'-(ß, γ-imido)triphosphate (GppNHp) indicated that state 1 is stabilized by intramolecular hydrogen-bonding interactions formed by Gln61. Since Ras are constitutively activated by substitution mutations of Gln61, here we determine crystal structures of the state 1 conformation of H-Ras•GppNHp carrying representative mutations Q61L and Q61H to observe the effect of the mutations. The results show that these mutations alter the mode of hydrogen-bonding interactions of the residue 61 with Switch II residues and induce conformational destabilization of the neighboring regions. In particular, Q61L mutation results in acquirement of state 2-like structural features. Moreover, the mutations are likely to impair an intramolecular structural communication between Switch I and Switch II. Molecular dynamics simulations starting from these structures support the above observations. These findings may give a new insight into the molecular mechanism underlying the aberrant activation of the Gln61 mutants.

Exploring ligand binding pathways on proteins using hypersound-accelerated molecular dynamics.

Capturing the dynamic processes of biomolecular systems in atomistic detail remains difficult despite recent experimental advances. Although molecular dynamics (MD) techniques enable atomic-level observations, simulations of "slow" biomolecular processes (with timescales longer than submilliseconds) are challenging because of current computer speed limitations. Therefore, we developed a method to accelerate MD simulations by high-frequency ultrasound perturbation. The binding events between the protein CDK2 and its small-molecule inhibitors were nearly undetectable in 100-ns conventional MD, but the method successfully accelerated their slow binding rates by up to 10-20 times. Hypersound-accelerated MD simulations revealed a variety of microscopic kinetic features of the inhibitors on the protein surface, such as the existence of different binding pathways to the active site. Moreover, the simulations allowed the estimation of the corresponding kinetic parameters and exploring other druggable pockets. This method can thus provide deeper insight into the microscopic interactions controlling biomolecular processes.

Microsecond-timescale MD simulation of EGFR minor mutation predicts the structural flexibility of EGFR kinase core that reflects EGFR inhibitor sensitivity.

Approximately 15-30% of patients with lung cancer harbor mutations in the EGFR gene. Major EGFR mutations (>90% of EGFR-mutated lung cancer) are highly sensitive to EGFR tyrosine kinase inhibitors (TKIs). Many uncommon EGFR mutations have been identified, but little is known regarding their characteristics, activation, and sensitivity to various EGFR-TKIs, including allosteric inhibitors. We encountered a case harboring an EGFR-L747P mutation, originally misdiagnosed with EGFR-del19 mutation using a routine diagnostic EGFR mutation test, which was resistant to EGFR-TKI gefitinib. Using this minor mutation and common EGFR-activating mutations, we performed the binding free energy calculations and microsecond-timescale molecular dynamic (MD) simulations, revealing that the L747P mutation considerably stabilizes the active conformation through a salt-bridge formation between K745 and E762. We further revealed why several EGFR inhibitors, including the allosteric inhibitor, were ineffective. Our computational structural analysis strategy would be beneficial for future drug development targeting the EGFR minor mutations.

E487K-Induced Disorder in Functionally Relevant Dynamics of Mitochondrial Aldehyde Dehydrogenase 2.

Mitochondrial aldehyde dehydrogenase 2 (ALDH2), which is a homotetramer assembled by two equivalent dimers, is an important enzyme that metabolizes ethanol-derived acetaldehyde to acetate in a coenzyme-dependent manner. The highly reactive acetaldehyde exhibits a toxic effect, indicating that the proper functioning of ALDH2 is essential to counteract aldehyde-associated diseases. It is known that the catalytic activity of ALDH2 is drastically impaired by a frequently observed mutation, E487K, in a dominant fashion. However, the molecular basis of the inactivation mechanism is elusive because of the complex nature of the dynamic behavior. Here, we performed microsecond-timescale molecular dynamics simulations of the proteins complexed with coenzymes. The E487K mutation elevated the conformational heterogeneity of the dimer interfaces, which are relatively distal from the substituted residue. Dynamic network analyses showed that Glu487 and the dimer interface were dynamically communicated, and the dynamic community further spanned throughout all of the subunits in the wild-type; however, this network was completely rearranged by the E487K mutation. The perturbation of the dynamic properties led to alterations of the global conformational motions and destabilization of the coenzyme binding required for receiving a proton from the catalytic nucleophile. The insights into the dynamic behavior of the dominant negative mutant in this work will provide clues to restore its function.

Landscape and function of multiple mutations within individual oncogenes.

Sporadic reports have described cancer cases in which multiple driver mutations (MMs) occur in the same oncogene1,2. However, the overall landscape and relevance of MMs remain elusive. Here we carried out a pan-cancer analysis of 60,954 cancer samples, and identified 14 pan-cancer and 6 cancer-type-specific oncogenes in which MMs occur more frequently than expected: 9% of samples with at least one mutation in these genes harboured MMs. In various oncogenes, MMs are preferentially present in cis and show markedly different mutational patterns compared with single mutations in terms of type (missense mutations versus in-frame indels), position and amino-acid substitution, suggesting a cis-acting effect on mutational selection. MMs show an overrepresentation of functionally weak, infrequent mutations, which confer enhanced oncogenicity in combination. Cells with MMs in the PIK3CA and NOTCH1 genes exhibit stronger dependencies on the mutated genes themselves, enhanced downstream signalling activation and/or greater sensitivity to inhibitory drugs than those with single mutations. Together oncogenic MMs are a relatively common driver event, providing the underlying mechanism for clonal selection of suboptimal mutations that are individually rare but collectively account for a substantial proportion of oncogenic mutations.

Improvement in predicting drug sensitivity changes associated with protein mutations using a molecular dynamics based alchemical mutation method.

While molecular-targeted drugs have demonstrated strong therapeutic efficacy against diverse diseases such as cancer and infection, the appearance of drug resistance associated with genetic variations in individual patients or pathogens has severely limited their clinical efficacy. Therefore, precision medicine approaches based on the personal genomic background provide promising strategies to enhance the effectiveness of molecular-targeted therapies. However, identifying drug resistance mutations in individuals by combining DNA sequencing and in vitro analyses is generally time consuming and costly. In contrast, in silico computation of protein-drug binding free energies allows for the rapid prediction of drug sensitivity changes associated with specific genetic mutations. Although conventional alchemical free energy computation methods have been used to quantify mutation-induced drug sensitivity changes in some protein targets, these methods are often adversely affected by free energy convergence. In this paper, we demonstrate significant improvements in prediction performance and free energy convergence by employing an alchemical mutation protocol, MutationFEP, which directly estimates binding free energy differences associated with protein mutations in three types of a protein and drug system. The superior performance of MutationFEP appears to be attributable to its more-moderate perturbation scheme. Therefore, this study provides a deeper level of insight into computer-assisted precision medicine.

Molecular Basis for Allosteric Inhibition of GTP-Bound H-Ras Protein by a Small-Molecule Compound Carrying a Naphthalene Ring.

The ras oncogene products (H-Ras, K-Ras, and N-Ras) have been regarded as some of the most promising targets for anticancer drug discovery because their activating mutations are frequently found in human cancers. Nonetheless, molecular targeted therapy for them is currently unavailable. Here, we report the discovery of a small-molecule compound carrying a naphthalene ring, named KBFM123, which binds to the GTP-bound form of H-Ras. The solution structure of its complex with the guanosine 5'-(ß,γ-imide) triphosphate-bound form of H-RasT35S (H-RasT35S·GppNHp) indicates that the naphthalene ring of KBFM123 interacts directly with a hydrophobic pocket located between switch I and switch II and allosterically inhibits the effector interaction by inducing conformational changes in switch I and its flanking region in strand ß2, which are directly involved in recognition of the effector molecules, including c-Raf-1. In particular, Asp38 of H-Ras, a crucial residue for the interaction with c-Raf-1 via the formation of a salt bridge with Arg89 of the Ras-binding domain (RBD) of c-Raf-1, shows a drastic conformational change: its side chain orients toward the opposite direction. Consistent with these results, KBFM123 exhibits an activity to inhibit, albeit weakly, the association of H-RasG12V·GppNHp with the c-Raf-1 RBD. The binding of the naphthalene ring to the hydrophobic pocket of H-RasT35S·GppNHp is further supported by nuclear magnetic resonance analyses showing that two other naphthalene-containing compounds with distinct structures also exhibit similar binding properties with KBFM123. These results indicate that the naphthalene ring could become a promising scaffold for the development of Ras inhibitors.

Structural transition of solvated H-Ras/GTP revealed by molecular dynamics simulation and local network entropy.

The state transitions of solvated H-Ras protein with GTP were theoretically analyzed through molecular dynamics (MD) simulations. To accelerate the structural changes associated with the locations of two switch regions (I and II), the Parallel Cascade Selection MD (PaCS-MD) method was employed in this study. The interconversions between the State 1 and State 2 were thus studied in atomic details, leading to a reasonable agreement with experimental observations and consequent scenarios concerning the transition mechanism that would be essential for the development of Ras inhibitors as anti-cancer agents. Furthermore, the state-transition-based local network entropy (SNE) was calculated for the transition process from State 1 to State 2, by which the temporal evolution of information entropy associated with the dynamical behavior of hydrogen bond network composed of hydration water molecules was described. The calculated results of SNE thus proved to provide a good indicator to detect the dynamical state transition of solvated Ras protein system (and probably more general systems) from a viewpoint of nonequilibrium statistical thermodynamics.

Structural basis for intramolecular interaction of post-translationally modified H-Ras•GTP prepared by protein ligation.

Ras undergoes post-translational modifications including farnesylation, proteolysis, and carboxymethylation at the C terminus, which are necessary for membrane recruitment and effector recognition. Full activation of c-Raf-1 requires cooperative interaction of the farnesylated C terminus and the activator region of Ras with its cysteine-rich domain (CRD). However, the molecular basis for this interaction remains unclear because of difficulties in preparing modified Ras in amounts sufficient for structural studies. Here, we use Sortase A-catalyzed protein ligation to prepare modified Ras in sufficient amounts for NMR and X-ray crystallographic analyses. The results show that the farnesylated C terminus establishes an intramolecular interaction with the catalytic domain and brings the farnesyl moiety to the proximity of the activator region, which may be responsible for their cooperative recognition of c-Raf-1-CRD.

Molecular Mechanism for Conformational Dynamics of Ras·GTP Elucidated from In-Situ Structural Transition in Crystal.

Ras•GTP adopts two interconverting conformational states, state 1 and state 2, corresponding to inactive and active forms, respectively. However, analysis of the mechanism for state transition was hampered by the lack of the structural information on wild-type Ras state 1 despite its fundamental nature conserved in the Ras superfamily. Here we solve two new crystal structures of wild-type H-Ras, corresponding to state 1 and state 2. The state 2 structure seems to represent an intermediate of state transition and, intriguingly, the state 1 crystal is successfully derived from this state 2 crystal by regulating the surrounding humidity. Structural comparison enables us to infer the molecular mechanism for state transition, during which a wide range of hydrogen-bonding networks across Switch I, Switch II and the α3-helix interdependently undergo gross rearrangements, where fluctuation of Tyr32, translocation of Gln61, loss of the functional water molecules and positional shift of GTP play major roles. The NMR-based hydrogen/deuterium exchange experiments also support this transition mechanism. Moreover, the unveiled structural features together with the results of the biochemical study provide a new insight into the physiological role of state 1 as a stable pool of Ras•GTP in the GDP/GTP cycle of Ras.