BiComp-DTA: Drug-target binding affinity prediction through complementary biological-related and compression-based...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Drug-target binding affinity prediction is crucial for early drug discovery.
• Existing experimental and data-driven methods have limitations, including reliance on scarce data or computational overhead from complex models.
• Current deep learning approaches often require extensive data and computational resources for protein and drug encoding.

Purpose of the Study:

• To develop an efficient and accurate method for drug-target binding affinity prediction.
• To propose a unified protein sequence encoding measure, BiComp, that captures compression-based and evolutionary features.
• To introduce BiComp-DTA, a deep neural network utilizing BiComp for enhanced prediction.

Main Methods:

• Developed BiComp, a unified measure for protein sequence encoding, integrating Normalized Compression Distance and Smith-Waterman algorithms.
• Utilized BiComp to encode protein sequences for a new deep neural network, BiComp-DTA.
• Evaluated BiComp-DTA on four benchmark datasets for drug-target binding affinity prediction.

Main Results:

• BiComp-DTA demonstrated superior accuracy, runtime efficiency, and fewer trainable parameters compared to state-of-the-art methods.
• The BiComp encoding proved more effective than its individual components (Normalized Compression Distance and Smith-Waterman) for binding affinity prediction.
• BiComp-DTA can be efficiently executed on standard desktop computers.

Conclusions:

• BiComp offers an efficient and effective protein sequence representation for drug-target binding affinity prediction.
• The proposed BiComp-DTA method provides a computationally efficient and accurate alternative to complex existing models.
• This approach reduces the need for extensive domain knowledge and multiple data sources in drug discovery.