TeM-DTBA: time-efficient drug target binding affinity prediction using multiple modalities with Lasso feature...

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug discovery relies on understanding drug-target interactions (DTIs).
• Computational methods, particularly deep learning, are increasingly used for DTI prediction due to cost and time constraints of wet-lab experiments.
• Current computational DTI prediction often treats the problem as binary classification, neglecting quantitative binding affinity, and can be computationally intensive.

Purpose of the Study:

• To develop a novel, time-efficient computational method for predicting drug-target binding affinity.
• To address the limitations of existing methods by incorporating quantitative binding affinity and optimizing computational efficiency.
• To accelerate virtual screening and drug repositioning processes.

Main Methods:

• Introduction of Time-efficient Multimodal Drug Target Binding Affinity (TeM-DTBA) method.
• Fusion of multimodal data including compound structures and target sequences.
• Application of Lasso feature selection to reduce dimensionality and improve training speed.

Main Results:

• The TeM-DTBA method demonstrates superior performance on benchmark datasets (KIBA and Davis).
• Achieved mean squared errors of 18.8% on KIBA and 23.19% on Davis datasets.
• Model training time was reduced by over 50% due to feature selection.

Conclusions:

• The TeM-DTBA method accurately predicts drug-target binding affinity.
• The approach offers a significant improvement in both accuracy and computational efficiency over state-of-the-art methods.
• This method can accelerate drug discovery pipelines, including virtual screening and drug repositioning.