#29 · Structural bio & protein design
Structure-aware deep learning predicts protein-ligand and compound-protein binding
The dominant thread is explicit three-dimensional structural information combined with sequence, language-model, or topological representations to predict molecular binding. HelixDTA couples sequence features with complete target structures for drug-target affinity on Davis, and MultiRSF fuses molecular surface fingerprints, RNA language model embeddings, and secondary structure to predict RNA-small-molecule binding sites. Persistent local Laplacian descriptors feed ML models for protein-ligand affinity on PDBbind benchmarks. DESTIN applies a fine-tuned 3D foundation model genome-wide to compound-protein interactions, while Seq2Pocket fine-tunes protein language models with structure-aware post-processing and FunFOLD5 combines local Foldseek template search with global fold similarity. The newer elements are topology-derived descriptors and 3D foundation-model representations at genome scale. Reported results are benchmark performance.
Published papers
HelixDTA: Dual-Branch Sequence-Structure Learning with Complete Target Structures for Robust and Interpretable Drug-Target Affinity Prediction.
HelixDTA is a dual-branch deep learning model combining sequence and complete target structure features for drug-target affinity prediction, tested on Davis…
MultiRSF: A Deep Learning Approach for Predicting RNA-Small-Molecule Binding Sites Using Surface Characteristics.
MultiRSF, a multimodal Transformer framework fusing molecular surface fingerprints, RNA language model embeddings, and secondary structure to predict…
Persistent Local Laplacian Prediction of Protein-Ligand Binding Affinities.
Introduces persistent local Laplacian molecular descriptors fed to ML models for protein-ligand binding affinity, benchmarked on PDBbind v2007, v2013, and…
Genome-wide-scale prediction of compound-protein interactions using foundation and language models based on three-dimensional structures of compounds and proteins.
DESTIN: an ML framework predicting compound-protein interactions genome-wide from 3D compound and protein structures, using a fine-tuned 3D foundation model…
FunFOLD5: combining localised templates and structural homology for accurate predictions of biologically relevant protein-ligand interactions.
FunFOLD5 is a template-based protein-ligand binding-site predictor combining local Foldseek template search with global fold similarity and BioLiP2…
Seq2Pocket: augmenting protein language models for spatially consistent binding site prediction.
Seq2Pocket fine-tunes protein language models with an extended sc-PDB training set and structure-aware post-processing to produce less fragmented ligand…
Topic history
| Date | Rank |
|---|---|
| #29 | |
| #19 | |
| #24 | |
| #15 | |
| #20 | |
| #15 | |
| #9 | |
| #30 |