Computational Drug Discovery Methods

Learning a chemistry-aware latent space for molecular encoding and generation with a large-scale Transformer Variational Autoencoder

The paper introduces a Transformer-based VAE for molecular encoding and generation, achieving high reconstruction accuracy and a well-structured latent space aligned with chemical similarity.

Pillbox: A Leakage-Aware Foundation-Model Predictor and Lineage-Ceiling Diagnostic for Cancer Drug Response

Pillbox predicts cancer drug response using foundation models and evaluates lineage ceiling through residual correlation analysis.

Systematic AI-Driven Drug Repurposing via Clinical Trial Data Mining: A Framework and Six Cross-Therapeutic Case Studies.

This paper introduces an AI tool that mines clinical trial data to identify drug repurposing candidates across six disease areas, finding new therapeutic potentials for existing drugs.

Virtual phenotypic screening discovers novel scaffolds inhibiting the PI3K/mTOR pathway

The paper introduces PhenoCompass, a model for virtual phenotypic screening that identifies novel PI3K/mTOR inhibitors from a large compound library.

ADMETron: An AI-driven SaaS platform for comprehensive ADMET prediction and compound prioritisation

ADMETron is an AI platform that predicts ADMET properties and prioritizes compounds using ensemble models, achieving top rankings on benchmark tests and offering advanced visualization tools.

Structure-Based Pharmacophore Modeling, High-throughput Screening, and Molecular Dynamics Identify a Novel DrugBank-Derived HIV-1 Protease Inhibitor

The paper identifies BAN as a novel HIV-1 protease inhibitor with binding affinity comparable to 3TL but through a distinct mechanism involving stronger polar interactions and increased protein flexibility.

AlphaFold3 for Structure-guided Ligand Discovery

Retrospective and prospective benchmarks show AlphaFold3 co-folding memorizes training poses and achieves half the prospective hit rate of DOCK3 at 6,700-fold higher cost, but excels as a post-docking filter.

seekrflow: Towards end-to-end automated simulation pipeline with machine-learned force fields for accelerated drug-target kinetic and thermodynamic predictions

seekrflow automates drug-target kinetic and thermodynamic predictions using a multiscale milestoning simulation pipeline, achieving accurate results with reduced manual intervention and computational cost.

Back to basics: Observed statistics are sufficient to predict drug responses

The paper introduces Rhaister, a perturbation-response predictor that uses summary statistics to predict drug responses, outperforming virtual-cell models in speed and accuracy.

Combinatorial docking and molecular generation to navigate over 100-billion molecules for prospective ligand discovery

The paper introduces CombiDOCK and MINT-Dock to efficiently screen large molecule libraries for ligand discovery, achieving high hit rates and computational efficiency.

AI-Enforced Ultra-Large Virtual Screening Discovers Potent CD28 Binders

This study uses AI-enhanced virtual screening to identify potent CD28 binders from a large chemical library, achieving submicromolar affinity and functional blockade.

A Deep Hypergraph Learning Model for Predicting Antimicrobial Combination Effects Across Bacterial Targets

This paper introduces a hypergraph neural network to predict antimicrobial combination effects across bacterial targets, achieving high accuracy and F1-scores in classifying synergy, antagonism, and non-interaction.

GeroQubit: a lightweight, honesty-first de-novo design platform for geroscience-native small molecules with calibrated uncertainty

GeroQubit is a de-novo molecule design platform that reports calibrated uncertainty and weak signals for geroscience targets, showing high performance on known compounds but poor performance on novel chemotypes.

Machine Learning-Guided Discovery of Bacterial-Selective Membrane-Active Compounds Reveals Mechanistic Bias in Antibiotic Training Datasets

This paper develops a machine learning pipeline to discover novel antibacterial compounds with distinct mechanisms of action, finding that training data bias limits the discovery of mechanistically novel antibiotics.

Calibrated Uncertainty Quantification for Patient-Level AML Drug Sensitivity Prediction Using Split Conformal Prediction

This paper applies split conformal prediction to patient-level AML drug response modeling, achieving well-calibrated uncertainty estimates across multiple confidence levels.

TITAN-BBB: Predicting BBB Permeability using Multi-Modal Deep-Learning Models

TITAN-BBB predicts BBB permeability using multi-modal deep learning, achieving high accuracy and error reduction compared to existing methods.

MolDeBERTa: Foundational Model for Physicochemical and Substructure-Informed Molecular Representation Learning

MolDeBERTa is a molecular language model pre-trained on SMILES data with novel objectives to improve property prediction and substructure similarity, outperforming existing models on multiple benchmarks.

APOSM: Pairwise preference learning improves generative small-molecule design

APOSM improves generative small-molecule design by using pairwise preference learning to enhance surrogate model reliability and active candidate selection.

Multi-level, multi-body atomic interaction graphs for machine learning-based prediction of protein-ligand binding energies

This paper introduces a graph-based scoring function for predicting protein-ligand binding energies, achieving high accuracy across multiple datasets.

ExplainBind: Explainable Physicochemical Determinants of Protein-Ligand Binding via Non-Covalent Interactions

ExplainBind predicts protein-ligand binding likelihood and identifies key residues and non-covalent interactions without structural inputs, outperforming existing methods and enabling drug discovery insights.

Stereochemistry-Aware Drug-Target Affinity Prediction

This paper introduces a stereochemistry-aware framework for drug-target affinity prediction, showing improved accuracy in distinguishing enantiomer affinities.

Targeting an allosteric binding site in the citrate transporter NaCT (SLC13A5)

This study identifies small molecule inhibitors of the NaCT transporter by targeting an allosteric site, revealing a chemically distinct series with improved potency.

A Drug-Target Specificity Foundation Model for Off-target Prediction, Repurposing, and Generative Design

This paper introduces dtSFM, a foundation model for drug-target interactions that predicts off-targets, identifies drug candidates, and generates novel molecules using thermodynamics-based sequence modeling.

Do Larger Models Really Win in Drug Discovery?A Benchmark Assessment of Model Scaling in AI-Driven Molecular Property and Activity Prediction

The paper benchmarks model scaling in AI-driven drug discovery, finding that compact models often outperform larger ones across various molecular property and activity prediction tasks.

An Agentic Platform for Drug Repurposing Unified across Molecular, Phenotypic, and Clinical Scales

LinkD is a drug repurposing platform that integrates molecular, phenotypic, and clinical data to prioritize drugs, showing reduced prostate cancer incidence with beta-blockers.

Structure-guided compound prioritization strategy for virtual screening identifies putative binders for the nuclear receptor LRH-1

The paper develops a compound prioritization strategy using a multi-layer perceptron trained on docked poses from multiple docking approaches to identify putative binders for the nuclear receptor LRH-1.

Prediction variability in physiologically based pharmacokinetic modeling of tissue disposition under deep uncertainty

The paper evaluates prediction variability in PBPK models under deep uncertainty, showing significant variation in PK statistics and model disagreement for certain chemical compounds.

Mapping Chemical Diversity: Descriptor-Guided Clustering of Natural Products in the COCONUT Database

The paper develops a framework to map natural product chemical space using descriptor-guided clustering in the COCONUT database, identifying optimal descriptors and clustering methods to characterize chemical diversity.

NetPolicy-RL: Network-Informed Offline Reinforcement Learning for Pharmacogenomic Drug Prioritization

NetPolicy-RL improves drug prioritization by integrating network biology with reinforcement learning to optimize ranking quality in pharmacogenomic screens.

Computational Analysis of ELOVL6 Structure and Inhibition for Rational Drug Design

This study uses computational methods to analyze ELOVL6 structure and inhibition, identifying key residues and binding pathways for drug design.