contrastive learning

PAMP translates turnover predictions into specific enzyme mutations, achieving high positive candidate rates and cumulative gains through gradient-guided projection.

This paper introduces hTSMNet, a method for zero-shot EEG decoding across datasets and paradigms using geometry-conditioned hyperalignment and Riemannian SPD manifold learning.

BacLLM improves protein sequence representations for AMR analysis by integrating functional knowledge, enhancing resistance-gene identification and mechanism prediction.

FedEpitope uses federated learning with LoRA to predict linear B-cell epitopes without centralizing data, achieving good performance across multiple clients.

This paper introduces a contrastive graph neural network to discover novel larvicidal compounds against Aedes aegypti, identifying LC-79 as a potent and selective candidate.

This paper improves cryo-EM reconstruction of small complexes using 2D template matching and high-resolution structures as priors, enhancing density of ligand-binding sites in a 43 kDa protein kinase.

CRISCross improves CRISPR-Cas9 off-target prediction by integrating epigenetic features with sequence data across cell types, showing better performance than sequence-only models.

This paper evaluates a heterogeneous GNN for AMR prediction from MALDI-TOF spectra, showing it performs better than MLP only at high spectral similarity.

AC-Diff improves liver MRI harmonization by separating anatomy from contrast, achieving better segmentation results.

RNASeek is a generative foundation model for RNA sequence design and functional prediction, using reinforcement learning to optimize RNA properties like stability and cleavage activity.

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.

CSGDA predicts drug responses across biological heterogeneities by addressing distribution shifts using a cell state-guided graph domain adaptation framework.

This paper evaluates DrugCLIP's contrastive learning embeddings for virtual screening, showing strong performance but limitations under realistic conditions.

This study evaluates contrastive learning-based approaches for activity cliff prediction in drug discovery, finding they improve embedding separation but have dataset-dependent accuracy impacts.

The paper calibrates drug target affinity predictions to align with IC50 values using similarity-guided methods, reducing median fold error across kinase and non-kinase panels.

ReTIF improves RNA-compound interaction prediction and binding-site localization by using task-specific interaction representations, outperforming baselines in multiple evaluation scenarios.

BindScreen improves virtual screening by using a protein-centric contrastive learning approach with asymmetric loss and batch design, achieving better performance with fewer computational resources.

Ptarmigan-1 uses contrastive learning to predict small molecule-protein interactions without requiring 3D structure, outperforming traditional methods on challenging targets.

CALFP-MHC improves peptide-MHC binding prediction by using chemically grounded fingerprints and contrastive learning, achieving high AUC and PPV across diverse alleles and class imbalance.

This paper introduces a 3D CNN with a trainable CWT layer and BiCL pre-training for sub-second seizure detection, achieving >95% accuracy on EEG data.

This study evaluates computational models for predicting pharmacogenomic variant effects, finding that StructureDCA outperforms other methods in distinguishing functional variants.

This paper develops interpretable adverse event vector representations using contrastive learning for pharmacovigilance, outperforming existing methods in drug-event causality prediction.

ConfDock provides atom-specific uncertainty quantification for molecular docking using conformal prediction, achieving narrower intervals while maintaining coverage guarantees across diverse protein-ligand systems.

CAFE improves co-folding models for discovering allosteric and cryptic binding sites by using orthosteric blockers to guide fragment exploration.

MolMAE is a multimodal masked autoencoder that learns molecular representations by integrating surface point clouds, 3D graphs, and SMILES-derived tokens, achieving strong performance on molecular property prediction tasks.

This paper introduces a method using two neural networks to design drug-binding proteins with high affinity, surpassing current state-of-the-art methods.

Nevermore improves lead optimization by balancing binding affinity and developability constraints using a database-grounded framework with interpretable molecular edits.

The paper introduces Specificity Foundation Models (SFM) that predict molecular binding across six domains using a physics-inspired architecture, achieving high retrieval accuracy without structural data.

TESSERA is a foundation model for the cancer genome that uses self-supervised learning to predict variant pathogenicity, tumour typing, and treatment response, revealing interpretable biomarkers for chemotherapy selection.

The paper introduces ProteomeLM-HPI, a language model adapted to predict cross-species host-pathogen interactions using proteome context and fine-tuning.