Exscalate contributes to publications and journals

Transferable biomolecular force fields are developed by fitting either ab initio or experimental data related to representative molecules and can then be used to model chemical entities that are similar to the ones they were developed for. However, once parametrized on a given dataset, they are difficult to refit once new chemical entities, sensing schemes, or functional forms are introduced. On the other hand, Machine Learning Force Fields (MLFF) have recently gained attention for their accuracy and ease of expanding their Applicability Domain (AD). Nonetheless, their prediction times make them incompatible with High-Throughput Virtual Screening (HTVS) requirements. In this work, we follow the inverse of the widely adopted approach with transferable force fields and propose a new condensation approach that takes advantage of machine learning algorithms to massively predict force field parameters. The generated numerical distributions are then condensed in a single value that captures in a statistical way the chemical variability of the underlying molecules sharing that specific force field parameter and giving rise to the distribution itself, improving 30× computational efficiency with limited reduction in predicted molecular geometries accuracy. When tested on the public release of the OpenFF Industry Benchmark Season 1 v1.1 dataset, the molecular structures optimized by minimizing the Potential Energy Surface built with condensed FF parameters only show a minor decrease in Root Mean Squared Deviation (RMSD) and Torsion Fingerprint Deviations (TFD) performances compared to those obtained using molecule-specific FF parameters predicted at runtime. To give more context, the original MLFF and its condensed version are evaluated with respect to several well-known transferable force fields widely used for biomolecular simulations.

The prediction of drugs metabolism by in silico techniques is gaining a growing interest due to the possibility to process large datasets allowing the stability and safety of new drug candidates to be evaluated during the early stages of the drug discovery process. To date, in silico models for metabolism prediction mainly exploits the ligand-based (LB) properties of the training molecules to predict the occurrence of a given metabolic reaction and/or the reactive site involved in the biotransformation. However, recent reports highlighted that structure-based (SB) modeling can be conveniently integrated with LB methods for drug metabolism prediction purpose, with the advantages to predict if a given molecule can fit the enzyme active site and which moiety approaches the catalytic residues. Herein, we developed machine learning models for UDP-glucuronosyltransferase (UGT)-mediated metabolism by using both LB and SB methods. In particular, this study was focused on UGT2B7 and UGT2B15 isoforms which are involved in the clearance of many drugs as well as in clinically relevant drug-drug interactions. First, molecular dynamics (MD) and docking simulations were combined to explore the binding mechanism of cofactor and substrate within the catalytic pocket of the studied UGT isoforms exploiting their AlphaFold structures. The analysis of the MD trajectories allowed an appropriate conformation of both UGT isoforms to be identified for the development of binary classification models. For this purpose, Random Forest algorithm and the metabolic data extracted from the MetaQSAR database were used. SB models were trained on a set of scoring functions and protein-ligand interaction fingerprints derived from docking, while the LB models were built on a set of physicochemical and constitutional descriptors. When the single models were evaluated, the LB classifiers outperformed the SB models. However, the application of a consensus strategy led to an improvement of the prediction accuracy if compared to the individual models, highlighting that LB and SB approaches convey complementary information whose aggregation allowed us to achieve better predictions than the single models.

Structure-based virtual screening approaches like molecular docking rely on accurately identifying and precisely calculating binding pockets to efficiently search for potential ligands. In this paper, we introduce GENEOnet, a machine learning model designed for volumetric protein pocket detection that employs Group Equivariant Non-Expansive Operators (GENEOs). These operators simplify model complexity and enable more informed domain knowledge integration by selecting specific physical and chemical properties for each operator to focus on, as well as how they should react. Unlike other methods in this field, GENEOnet has fewer model parameters, resulting in reduced training costs, and offers greater explainability, allowing the parameters to be easily interpreted. GENEOnet processes the empty space within a protein by converting it into a 3D grid of uniform blocks, known as ‘voxels’. It then identifies regions of the grid with an output value above a threshold, thus producing a list of predicted pockets, ranked according to the model’s average output value. Our experimental results show that GENEOnet performs robustly even with small training datasets of 200 proteins and surpasses other established state-of-the-art methods in various metrics. Specifically, GENEOnet’s H1 score indicating the probability that the top-ranked pocket is the correct one is 0.764, compared to 0.702 for P2Rank, the next best performing algorithm on our PDBbind test set. Moreover, a case study considering various ABL1 kinase conformations demonstrates the excellent agreement between GENEOnet’s predictions and experimental sites. GENEOnet is available as a web service at https://geneonet.exscalate.eu, where users can access the pre-trained model for detecting and ranking protein cavities.

Molecular docking is an essential step in the drug discovery process involving the detection of three-dimensional poses of a ligand inside the active site of the protein. In this paper, we address the Molecular Docking search phase by formulating the problem in quadratic unconstrained binary optimization terms, suitable for an annealing approach. We propose a problem formulation as a weighted subgraph isomorphism between the ligand graph and the grid of the target protein pocket. In particular, we applied a graph representation to the ligand embedding all the geometrical properties of the molecule including its flexibility, and we created a weighted spatial grid to the 3D space region inside the pocket. The proposed quantum annealing-based method for molecular docking achieves valid ligand placements. Compared to simulated annealing, quantum solvers sampled fewer but higher-quality solutions with lower root-mean-square deviation, demonstrating competitive performance within hardware limits.

Since its emergence in late 2019, SARS-CoV-2, the causative agent of COVID-19, has continued to spread globally, with more than 7 million reported deaths as of March 2025. Among the viral nonstructural proteins, nsp12 serves as the RNA-dependent RNA polymerase (RdRp), mediating viral genome replication and transcription in concert with its cofactors nsp7 and nsp8. To date, only two nucleoside analogs specifically targeting SARS-CoV-2 nsp12, remdesivir and molnupiravir, have been authorized by the FDA for COVID-19 treatment. In response to the need for additional safe and effective antiviral agents, we screened two extensive in silico libraries of safe-in-man compounds (>9,000) and natural compounds (>249,000), against the SARS-CoV-2 nsp12/7/8 complex, targeting the orthosteric and two allosteric nsp12 sites, using the EXSCALATE (EXaSCale smArt pLatform Against paThogEns) platform. Compounds were then selected based on docking score significance, novelty for the target, and clinical safety profiles. The top 119 candidates were subsequently evaluated in a biochemical assay to assess their potential to inhibit SARS-CoV-2 nsp12/7/8 polymerase activity, identifying 42 compounds able to block it, among which four showed IC50 and EC50 values in the nanomolar or low micromolar range. When tested in cell-based assays to evaluate their efficacy on SARS-CoV-2 replication, they proved to inhibit it in the same concentration ranges. Mechanism of action studies revealed different modalities of inhibition. These results provide the basis for the development of novel antiviral compounds against SARS-CoV-2, targeting both the RdRp active site and an allosteric site, further suggesting that the Computer-Aided Drug Discovery (CADD) approach, together with experimental validation, can provide the basis for accelerated antiviral drug development.

Replacement of the phenolic hydroxy in 3-((1R,5S,9R)-2-phenethyl-9-vinyl-2-azabicyclo[3.3.1]nonan-5-yl)phenol (DC-1–76.2), a potent efficacious MOR agonist, with an amide bioisosteric moiety provided a MOR partial agonist with morphine-like potency in the forskolin-induced cAMP accumulation assay and in the [35S]GTPγS functional assay. This amide, 5, had superior metabolic stability in comparison to its precursor in human and mouse liver microsomes. However, in an antinociception study, an assay of pain-depressed locomotion in mice, it was found to possess shorter antinociceptive activity than its precursor. The in vitro and in vivo data enabled the characterization of amide, 5, as a functionally selective, low-efficacy, and low-potency MOR agonist with a relatively short duration of action in vivo. Modification of the N-phenethyl substituent in DC-1–76.2 gave a number of highly interesting partial agonists, and the unexpectedly potent antagonist, 17. The results of molecular docking and binding free energy calculations for DC-1–76.2 and 17 provided details about their receptor interactions and supported their functional roles. Several analogs synthesized were found to have sufficient potency in vitro to warrant further study.

Replacement of the phenolic hydroxy in 3-((1R,5S,9R)-2-phenethyl-9-vinyl-2-azabicyclo[3.3.1]nonan-5-yl)phenol (DC-1–76.2), a potent efficacious MOR agonist, with an amide bioisosteric moiety provided a MOR partial agonist with morphine-like potency in the forskolin-induced cAMP accumulation assay and in the [35S]GTPγS functional assay. This amide, 5, had superior metabolic stability in comparison to its precursor in human and mouse liver microsomes. However, in an antinociception study, an assay of pain-depressed locomotion in mice, it was found to possess shorter antinociceptive activity than its precursor. The in vitro and in vivo data enabled the characterization of amide, 5, as a functionally selective, low-efficacy, and low-potency MOR agonist with a relatively short duration of action in vivo. Modification of the N-phenethyl substituent in DC-1–76.2 gave a number of highly interesting partial agonists, and the unexpectedly potent antagonist, 17. The results of molecular docking and binding free energy calculations for DC-1–76.2 and 17 provided details about their receptor interactions and supported their functional roles. Several analogs synthesized were found to have sufficient potency in vitro to warrant further study.

Tryptase is a tetrameric serine protease and a key component of mast cell granules. Here, we explored an integrated approach to characterize tryptase ligands, combining novel experimental binding studies using Surface Plasmon Resonance, with in silico analysis through the Exscalate platform. For this, we focused on three inhibitors previously reported in the literature, including a bivalent inhibitor and its corresponding monovalent compound. All three ligands showed concentration-dependent binding to immobilized human tryptase with the bivalent inhibitor showing the highest affinity. Furthermore, Rmax values were similar, indicating that the compounds occupy all four binding pockets of the tryptase tetramer. This hypothesis was supported by in silico computational analysis that revealed the binding mode of the monovalent ligand, one in each monomer pocket, compared with crystal structure of the bivalent one, which simultaneously occupies two binding pockets. Additionally, we solved the 2.06 Å X-ray crystal structures of human Tryptase beta-2 (hTPSB2), in both its apo form and in complex with compound #1, experimentally confirming the binding mode and the key molecular interactions predicted by docking studies for this compound. This integrated approach offers a robust framework for elucidating both the strength and mode of interaction of potential tryptase inhibitors.

SARS-CoV-2 encodes a 3C-like protease (3CLpro) that is essential for viral replication. This cysteine protease cleaves viral polyproteins to release functional nonstructural proteins, making it a prime target for antiviral drug development. We investigated the inhibitory effects of halicin, a known c-Jun N-terminal kinase inhibitor, on 3CLpro. Mass spectrometry and crystallographic analysis revealed that halicin covalently binds to several cysteine residues in 3CLpro. As expected, Cys145, the catalytic residue, was found to be the most targeted residue by halicin. Secondly, Cys44 was found to be modified, suggesting a potential inhibitory role of this residue. A mutant protease (Cys44Ala) was generated to further understand the function of Cys44. In silico and enzymatic assays showed that the mutation significantly reduced the stability and activity of 3CLpro, indicating the importance of Cys44 in maintaining the active conformation of the protease. Differential scanning fluorimetry assays confirmed this evidence, showing a reduced thermal stability of the mutant compared to the wild-type protease. Our results highlight the potential of halicin as a multi-target inhibitor of 3CLpro and underline the importance of Cys44 in the function of the protease. These findings contribute to the development of effective antiviral therapies against COVID-19 by targeting critical residues in 3CLpro.

Successful virtual screening of therapeutically relevant proteins depends on selecting optimal protein structures and accurately identifying druggable binding sites. This study presents a novel methodology that combines pocket analysis, molecular docking, and molecular dynamics simulations to prioritize protein binding sites, using SARS-CoV-2 spike protein as a case study. The methodology begins by collecting a comprehensive dataset of resolved spike protein structures and known inhibitors. A pocket search identified potential binding sites across various protein conformations, followed by docking simulations to evaluate ligand-binding affinities. We introduce a novel algorithm COMPASS: COMputational Pocket Analysis and Scoring System includes the calculation of Pocket Frequency Score, which assesses pocket relevance based on the frequency of key residues, was introduced to refine pocket selection. This scoring system was combined with traditional pocket and docking scores to produce a Global Score, enabling the ranking of pockets. The top-ranked pockets underwent molecular dynamics simulations and free energy calculations to assess their stability and druggability. Six out of the ten best-ranked pockets demonstrated stable interactions with all tested inhibitors, highlighting their potential as drug targets. The study found that the selected pockets not only showed significant structural uniqueness but also correlated well with experimentally validated binding sites, confirming the method’s effectiveness. In conclusion, the proposed algorithm and the method enhance the accuracy of structure-based drug discovery by enabling the rational selection of protein-binding pockets. It shows potential for enhancing virtual screening efforts, especially for proteins with numerous available experimental structures, where selecting an optimal structure is critical.