Affinity2Vec: drug-target binding affinity prediction through representation learning, graph mining, and machine...

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Drug-target interaction (DTI) prediction is vital for drug repositioning and virtual screening.
• Existing regression methods for drug-target binding affinity (DTBA) often require 3D target structures, which are frequently unavailable.
• Non-structure-based approaches are needed to overcome 3D structure limitations in DTBA prediction.

Purpose of the Study:

• To propose Affinity2Vec, a novel regression-based method for DTBA prediction that does not require 3D structural information.
• To formulate the DTBA prediction task as a graph-based problem using a weighted heterogeneous graph.

Main Methods:

• Constructed a weighted heterogeneous graph integrating drug-drug similarity, target-target similarity, and drug-target binding affinities.
• Employed feature representation learning, graph mining, and machine learning techniques.
• Predicted drug-target binding affinity without relying on 3D structural data.

Main Results:

• Affinity2Vec demonstrated robustness and efficiency on benchmark datasets.
• The method achieved superior and competitive results compared to state-of-the-art non-structure-based DTBA prediction methods.
• Performance was evaluated using metrics such as mean squared error, rm2, concordance index, and area under the precision-recall curve.

Conclusions:

• Affinity2Vec provides an effective non-structure-based approach for drug-target binding affinity prediction.
• The graph-based formulation and integration of diverse data sources enhance prediction accuracy.
• This method has significant implications for accelerating drug discovery and repositioning efforts.