TEFDTA: a transformer encoder and fingerprint representation combined prediction method for bonded and non-bonded drug–target affinities

• DOI: 10.1093/bioinformatics/btad778
• Publication Date: 2023-12-23T20:42:05Z
• Authors: Zongquan Li; Pengxuan Ren; Hao Yang; Jie Zheng; Fang Bai
• Keywords: Affinities; Representation

Abstract Motivation The prediction of binding affinity between drug and target is crucial in discovery. However, the accuracy current methods still needs to be improved. On other hand, most deep learning focus only on non-covalent (non-bonded) molecular systems, but neglect cases covalent binding, which has gained increasing attention field development. Results In this work, a new attention-based model, A Transformer Encoder Fingerprint combined Prediction method for Drug–Target Affinity (TEFDTA) proposed predict bonded non-bonded drug–target interactions. To deal with such complicated problems, we used different representations protein molecules, respectively. detail, an initial framework was built by training our model using datasets protein–ligand For widely dataset Davis, additional contribution study that provide manually corrected Davis database. subsequently fine-tuned smaller interactions from CovalentInDB database optimize performance. results demonstrate significant improvement over existing approaches, average 7.6% predicting remarkable 62.9% compared BindingDB data alone. At end, potential ability identify activity cliffs investigated through case study. indicate sensitive discriminate difference affinities arising small variances structures compounds. Availability implementation codes TEFDTA are available at https://github.com/lizongquan01/TEFDTA.