AptaTrans: a deep neural network for predicting aptamer-protein interaction using pretrained encoders

Area of Science:

• Biomolecular engineering
• Computational biology
• Drug discovery

Background:

• Aptamers, DNA/RNA biomaterials, show promise for drug discovery.
• Systematic evolution of ligands by exponential enrichment (SELEX) identifies aptamers but faces time and accuracy limitations.
• Existing in silico methods for aptamer-protein interaction (API) prediction often neglect physicochemical interactions.

Purpose of the Study:

• To develop an accurate and efficient computational method for predicting aptamer-protein interactions (API).
• To enhance the drug discovery process by improving the efficiency of aptamer selection.

Main Methods:

• A deep learning pipeline, AptaTrans, was developed for API prediction.
• AptaTrans utilizes transformer-based encoders for monomer-level analysis of aptamer and protein sequences.
• Pretrained encoders were employed for structural representation, and the pipeline was validated on a benchmark dataset.

Main Results:

• AptaTrans demonstrated superior performance compared to existing models in predicting API.
• The pipeline's efficacy was confirmed through experimental validation.
• AptaTrans was integrated into a toolset with Apta-MCTS for aptamer candidate generation.

Conclusions:

• AptaTrans significantly improves the cost-effectiveness and efficiency of SELEX in drug discovery.
• The developed pipeline offers a more accurate approach to predicting API.
• Source code and dataset are publicly available for AptaTrans.