LDS-CNN: a deep learning framework for drug-target interactions prediction based on large-scale drug screening

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Drug-target interaction (DTI) is crucial for drug design and understanding complex diseases.
• Current challenges include vast protein data, high experimental costs, and handling diverse, incompatible data formats.
• A unified model for comprehensive DTI analysis is significantly needed.

Purpose of the Study:

• To propose a general and unified method for predicting drug-target interactions.
• To address the difficulties in analyzing large-scale and heterogeneous data formats in DTI studies.
• To develop a computational approach that reduces experimental costs and time.

Main Methods:

• A novel deep learning model, Large-scale Drug target Screening Convolutional Neural Network (LDS-CNN), was developed.
• LDS-CNN employs unified encoding to integrate and process different data formats.
• The method focuses on feature abstraction and prediction of potential drug-target interactions.

Main Results:

• LDS-CNN achieved high predictive performance on a large dataset (898,412 interactions, 1683 compounds, 14,350 proteins).
• Key metrics include an Area Under the Curve (AUC) of 0.96, Area Under the Precision-Recall Curve (AUPRC) of 0.95, and 90.13% accuracy.
• The model demonstrated effectiveness for large-scale datasets and data with varying formats.

Conclusions:

• The proposed LDS-CNN method provides a feasible solution for unified encoding of large-scale, multi-format DTI data.
• It efficiently abstracts features from diverse drug-related data, reducing experimental costs and time.
• This work offers a valuable reference for applying deep learning to DTI prediction and identifying potential drug candidates for complex diseases.