Computational Drug Discovery Methods

Hybrid Mechanistic-Neural Modeling of Concentration-Time Dynamics Generalizes Human Pharmacokinetics Prediction Across Unseen Chemical Space

PK-MUSE improves pharmacokinetics prediction by combining mechanistic models with neural corrections, achieving better performance across unseen chemical structures.

How many crystal structures do you need to trust your docking results?

The paper evaluates how many crystal structures are needed for accurate pose prediction in drug discovery, finding that five structures per scaffold often suffice.

Structure-based virtual screening reveals new metabolite interactions and modulators of human glucose-6-phosphate isomerase

This study identifies new metabolite modulators of human glucose-6-phosphate isomerase through structure-based virtual screening and experimental validation.

PredHLM: quantitative and interpretable prediction of metabolic half-life in human liver microsomes

PredHLM predicts metabolic half-life in human liver microsomes using a quantitative machine learning model, achieving RMSE 0.507 and R2 0.431 on a test set.

GRASP: Graph Representation Learning with Assay Supervision for Molecular Properties

GRASP is a graph Transformer that learns molecular structure from ZINC20 data and adapts to ChEMBL bioactivity, achieving better performance than existing models on ADMET prediction tasks.

Tractable and Thermodynamically Consistent pKa Prediction via First-Dissociation Aggregation

This paper introduces DTi-pKa, a graph neural network that predicts pKa values with thermodynamic consistency and achieves high accuracy on benchmark datasets.

From Benchmark to Bench: Can Agents Survive Real-World Drug Discovery?

The paper evaluates an agentic system for drug discovery, finding that its performance depends on predictive models rather than generation methods, and that outputs are indistinguishable from expert work.

Transcriptomic data and biomedical literature synergize in finding pharmacologic gene regulators

The paper introduces SNACKKSS, a tool that integrates RNA-Seq data and biomedical literature to predict protein-inhibiting drugs, showing improved performance when combined with existing predictors.

Bioactivity-Driven Prediction of Antibacterial Synergy Using Machine Learning Models

This paper develops a machine learning framework to predict antibacterial synergy using multi-level bioactivity features and demonstrates its performance under rigorous cross-validation.

Robust prediction of drug combination side effects in realistic settings

DCSE predicts drug combination side effects using a deep learning model, outperforming existing methods in realistic settings with limited data.

Integrating Developmental Neurotoxicity Prediction with Adverse Outcome Pathway Reasoning

This study integrates large language models with adverse outcome pathways to predict developmental neurotoxicity and generate mechanistic explanations, showing improved performance with few-shot prompting.

DESPOT: Direction-Enhanced Scoring POTentials

DESPOT improves protein-ligand docking by introducing anisotropic scoring potentials that outperform isotropic methods in docking accuracy and virtual screening.

Pan-Screening the Structural Predictability of Drug Toxicity: Validation of Molecular Fingerprints

This paper evaluates the predictability of drug structures against adverse reactions using molecular fingerprints, identifies high-confidence toxicities, and demonstrates the effectiveness of MACCS fingerprints for safety screening.

Reaction-aware hypergraph learning of chemical representations with LARK

This paper introduces LARK, a framework that learns chemical representations from reactions, showing improved performance in reaction prediction and molecular property prediction.

A Dynamics-Informed Machine Learning Framework for Inhibitor Optimization Against Quickly Evolving Targets to Avoid Resistance

This paper introduces ROBUST, a machine learning framework that identifies energetic determinants of inhibitor potency against evolving targets, demonstrating its utility in predicting and optimizing inhibitors with improved resistance profiles.

In-silico Exploration of Drug Leads Targeting Mycobacterial Ag85C

This study identifies potential inhibitors of the Mycobacterium tuberculosis Ag85C enzyme using docking, pharmacophore modeling, and ADME analysis, highlighting three promising compounds with favorable drug-like properties.

HMVA-DDI: A Hierarchical Multi-View Attention Network for Drug-Drug Interaction and Severity Prediction

HMVA-DDI predicts drug-drug interactions and their severity using a two-stage deep learning framework with improved attention mechanisms and task-specific features.

A biological-response compound representation allows chemical perturbation prediction across cell lines

The paper introduces BioPert, a framework that uses gene expression responses to predict chemical perturbations across cell lines, achieving higher accuracy than chemical structure-based methods.

Regulon-informed cellular representations reveal task-dependent generalization in drug combination prediction

The paper evaluates how different biological representations affect drug combination prediction across varying cellular contexts and finds that regulons improve performance on some tasks but not all.

Biological rationales from language models enable leakage-resistant forecasts of target-indication success

PRIORITI uses language models to predict target-indication success in drug discovery, achieving high accuracy and robustness across different benchmarks.

Tri-Modality Representation Learning for Molecular Property Prediction

This paper introduces TMCA, a multimodal framework that integrates SMILES, molecular graphs, and 3D conformations for molecular property prediction, achieving the best performance on benchmark datasets.

DS5: An Open-Source Framework for Standardized High-Throughput Drug Screening Data Storage, Analysis, and Drug Prioritization

DS5 is an open-source framework that standardizes HTS data storage, analysis, and drug prioritization, validated against public benchmarks and applied in clinical settings.

Heterogeneous Graph Contrastive Learning for Drug-Gene-Disease Motif Prediction

HANAMI improves drug-gene-disease motif prediction by integrating heterogeneous biomedical data and using graph contrastive learning, achieving up to 6% better performance than existing methods.

MetExPred: A Comprehensive Prediction Framework with Protein-Context-Aware Multi-view Learning for Drug Metabolism and Excretion

MetExPred predicts drug metabolism and excretion by integrating molecular and protein representations, achieving strong performance across multiple endpoints.

AnewDDE: An Agentic Drug Discovery Engine for Biomolecular Interaction Modelling and Closed-Loop Design

AnewDDE integrates AI tools for biomolecular interaction modelling and drug design, demonstrating improved performance in structure prediction, binder identification, affinity estimation, and ADMET prediction.

A closed-loop language-model agent for target-specific multi-objective hit-to-lead optimization

This paper develops a language-model agent for hit-to-lead optimization that outperforms existing methods on four kinase targets in a computational benchmark.

Kryptix-1: Conformational Ensemble Sampling Substantially Improves Cryptic Pocket Detection Relative to Static-Structure and Prior Computational Approaches

Kryptix-1 improves cryptic pocket detection by combining conformational sampling with a consensus algorithm, outperforming static-structure methods and prior computational approaches on a benchmark dataset.

Boltz2-Notebook: An Interactive Google Colab Platform for Diffusion-Based Biomolecular Structure and Binding Affinity Prediction using the Boltz2 model.

Boltz2-Notebook provides an accessible Colab interface for using the Boltz2 model to predict biomolecular structures and binding affinities, with benchmarking of its performance on protein-ligand pairs.

CycDelta: Multimodal Residue-Graph Learning of Permeability Changes in Cyclic Peptides

CycDelta is a multimodal residue-graph model for predicting permeability changes in cyclic peptides by comparing within-source permeability differences, outperforming baselines on out-of-distribution and zero-shot tasks.

Deciphering Mechanistic Signatures in Drug-Drug Interactions with Dual Topology Graphs

This paper introduces DualTopoDDI, a deep learning model for predicting drug-drug interactions with mechanistic insights, achieving high accuracy and interpretability.