Baghdad Science Journal (Nov 2024)

Deep Learning based Models for Drug-Target Interactions

  • Ali K. Abdul Raheem,
  • Ban N. Dhannoon

DOI
https://doi.org/10.21123/bsj.2024.9212
Journal volume & issue
Vol. 21, no. 11

Abstract

Read online

The typical drug development approach is slow, costly, and fraught with failure - scientists examine millions of compounds, but only a few make it to preclinical or clinical testing. Machine learning (ML), a subset of AI, is a fast-expanding subject many pharmaceutical businesses increasingly utilize. Incorporating machine learning technologies into the drug development process can aid in automating repetitive data processing and analysis processes. ML techniques may be used at several stages of drug development, including drug property prediction, drug-target interaction (DTI) prediction, and De Novo drug design. DTIs are a critical component of the drug development process. When a drug (a chemical molecule) attaches to a target (proteins or nucleic acids), it is said to bind; it alters its biological behavior/function, returning it to normal. DTI prediction is an essential part of the Drug Discovery process since it may speed up and decrease costs, but it is challenging and costly because experimental assays take a long time and are expensive. In recent years, deep learning-based approaches have demonstrated encouraging results in predicting DTI. This paper developed two deep-learning architectures to predict drug-target interactions. The first model uses message-passing neural networks (MPNN) for drug encoding and bidirectional gated recurrent units (Bi-GRU) for protein-encoding. The second model uses Bi-GRU for drug encoding and protein encoding. The two models were trained and evaluated on several benchmark datasets. Our results demonstrate that our models outperform state-of-the-art DTI prediction methods and are a promising approach for predicting DTI with high accuracy.

Keywords