Computational Drug Discovery Methods

Plixer generates drug-like molecules for protein pockets using a two-stage model that decouples spatial interaction learning from chemical grammar learning.

DrugDL is a deep learning framework that improves drug molecule representation and multi-task prediction, outperforming existing methods in cold-start scenarios and aiding drug discovery.

The paper improves structure-based virtual screening by modifying AlphaFold2-predicted protein structures to better capture ligand-induced conformations, enhancing screening accuracy.

This paper develops small molecule inhibitors targeting Mtb cholesterol oxidases CYP125 and CYP142, showing potent antimicrobial activity against drug-susceptible and MDR-TB strains.

This paper improves ADMET prediction by augmenting Mol2Vec embeddings with classical descriptors and feature selection, achieving top results on 10 of 16 TDC benchmarks.

MGMG generates bioactivity-aware molecules using cell morphology and textual descriptions without target information, achieving comparable binding affinities to reference compounds.

SCOPE identifies molecular mechanisms in phenotypic screens by linking compound data to pathways and targets, revealing calcium signaling and HRH1 as key in exodosis regulation.

The paper designs small-molecule inhibitors of hIL-6 using computational screening and validates one compound's inhibitory effect on IL6-induced STAT3 activity.

This study identifies 10 phytochemicals from Cochlospermum species as potential HMGR inhibitors with binding affinities comparable to statins.

The paper evaluates the structural dynamics and interactions of LVDD9B protein with WSSV VP26 and identifies potential therapeutic compounds for antiviral therapy in aquaculture.

The paper develops a machine learning model to predict drug-induced phospholipidosis using a large dataset of repurposed drugs tested across multiple cell lines.

This paper introduces FilmCPI, a sequence-based model that leverages data imbalance to improve compound-protein interaction prediction, showing better performance and generalization across datasets.

DREAM-GNN improves drug repositioning by integrating heterogeneous biomedical data through a dual-route graph neural network framework, outperforming existing methods in recovering repositioning candidates.

This study uses network pharmacognosy to analyze the molecular interactions of Galphimia glauca bioactive compounds with human proteins, identifying key targets and pathways involved in its therapeutic effects.

The paper evaluates deep learning models for drug response prediction, finds transcriptomics data drives performance, and introduces new models that learn chemical features.

Top-DTI predicts drug-target interactions by combining topological features and large language model embeddings, outperforming existing methods on benchmark datasets.

SVM and logistic regression classifiers of Mpro inhibitor potency reach test ROC AUC 0.86 and 0.83; potency favours hydrophilic compounds, trading off against drug-likeness.

This study identifies antiviral compounds from traditional plants against HMPV using computational methods, finding Glycyrrhizin and Withaferin A as promising candidates with strong binding and pharmacokinetic properties.

This paper predicts protein-ligand binding affinity using classical and quantum-enhanced models, finding that hybrid quantum-classical models achieve high accuracy on curated datasets.

The paper introduces PATH, an interpretable algorithm for predicting protein-ligand binding affinity using persistent homology, achieving comparable performance with reduced computational complexity.

This study designs multifunctional peptides targeting collagen remodelling, anti-inflammation, and ECM stability using computational methods, identifying candidates with potential dual or pan-target applications.

This paper introduces Kinhibit, a framework that uses pre-trained molecular encoders and protein language models to predict kinase-inhibitor binding affinity, achieving high accuracy on MAPK pathway kinases and a large kinase dataset.

DrugTar improves druggability prediction by integrating sequence embeddings and gene ontologies, achieving high performance in target discovery.

RNAsmol is a sequence-based deep learning framework that predicts RNA-small molecule interactions using data perturbation and augmentation, achieving improved performance over existing methods.

MFPLI evaluates protein-ligand binding authenticity using deep learning with physicochemical and structural features, achieving high accuracy in distinguishing native conformations from docking poses.

This study evaluates four large language models for validating drug repurposing hypotheses using structured prompts and benchmark cases, finding GPT-4o and DeepSeek most reliable with improved accuracy and precision.

This study introduces ensemble docking methods to predict small molecule binding affinities and dynamics for intrinsically disordered proteins, showing agreement with experimental data.

This study benchmarks docking tools on experimental and AI-predicted protein structures to assess their accuracy in simulating protein-ligand interactions.

This paper combines crystallographic fragment screening, algorithmic merging, and automated synthesis to generate scaffolds targeting NCS-1 PPIs involved in neurological disorders.

This paper prioritizes potential drugs for repurposing by analyzing pathway-disease-drug relationships and network proximity using transcriptomic and PPI data.