3DICE: Interpretable 3D Cross-Modal Learning for Drug–Target Interaction Prediction
Published in Bioinformatics, 3DICE combines Uni-Mol and ESM-IF1 structural embeddings to model drug–protein relationships. It outperformed state-of-the-art models in accuracy, generalisability, and efficiency on DrugBank and KIBA, while highlighting influential atoms and protein residues for intrinsic interpretability.
Second Award in Computational Biology and Bioinformatics at Regeneron ISEF 2026.