Drug-target interaction prediction with collaborative contrastive learning and adaptive self-paced sampling strategy | JoVE Visualize

Area of Science:

Bioinformatics
Computational Chemistry
Drug Discovery

Background:

Drug-target interaction (DTI) prediction is crucial for identifying potential drug candidates in drug discovery and repositioning.
Existing methods often fail to fully leverage complementary relationships across multiple biological networks, hindering consistent representation learning.
The selection of negative samples critically impacts the performance of contrastive learning in DTI prediction.

Purpose of the Study:

To propose a novel deep learning model, CCL-ASPS, for enhanced drug-target interaction prediction.
To integrate Collaborative Contrastive Learning (CCL) and Adaptive Self-Paced Sampling (ASPS) to improve representation consistency and negative sample selection.
To address the limitations of previous DTI prediction approaches by utilizing multi-network information and dynamic sampling.

Main Methods:

Developed CCL-ASPS, a deep learning model combining Collaborative Contrastive Learning (CCL) and Adaptive Self-Paced Sampling (ASPS).
CCL enables learning fused embeddings of drugs and targets by leveraging multiple biological networks for consistent representations.
ASPS dynamically selects informative negative sample pairs to optimize contrastive learning.

Main Results:

CCL-ASPS demonstrated significant improvements in drug-target interaction prediction accuracy compared to state-of-the-art methods on established datasets.
Ablation experiments validated the effectiveness and contributions of both the Collaborative Contrastive Learning and Adaptive Self-Paced Sampling strategies.
The model achieved notable enhancements in predictive performance, confirming its efficacy.

Conclusions:

The proposed CCL-ASPS model effectively overcomes limitations of prior DTI prediction methods by integrating CCL and ASPS.
CCL-ASPS significantly improves DTI predictive performance, offering a more robust approach compared to existing techniques.
Case studies and cold start experiments highlight CCL-ASPS's capability in predicting novel drug-target interactions, aiding future drug discovery efforts.