DTITR: End-to-end drug-target binding affinity prediction with transformers

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate identification of Drug-Target Interactions (DTIs) is crucial for drug discovery.
• Current computational methods often use binary classification, neglecting binding strength and interaction details.
• Existing models lack interpretability and fail to capture complex binding mechanisms.

Purpose of the Study:

• To develop an end-to-end Transformer-based architecture for predicting drug-target binding affinity (DTA).
• To utilize 1D sequential and structural data for protein and compound representation.
• To improve the accuracy and interpretability of DTI predictions.

Main Methods:

• Proposed an end-to-end Transformer architecture utilizing self-attention and cross-attention layers.
• Represented proteins and compounds using 1D raw sequential and structural data.
• Employed Transformer-Encoders to generate robust aggregate representations for DTA prediction.

Main Results:

• The Transformer-based model achieved superior performance in predicting DTA compared to state-of-the-art baselines.
• The model accurately predicted interaction strength values and discriminated the rank order of binding affinity.
• The inclusion of a Cross-Attention Transformer-Encoder significantly enhanced the model's discriminative power.

Conclusions:

• The developed Transformer architecture is effective for drug-target binding affinity prediction.
• The model offers improved accuracy and interpretability in drug discovery.
• This approach validates the use of attention mechanisms for understanding complex DTIs.