Drug Discovery: Protein-Protein Interaction Prediction using Decoder-only Transformer Model

Understanding protein-protein interactions (PPIs) are essential for comprehending various biological processes, including disease causing microbes pathological mechanisms. With the rapid advancement in deep learning (DL) techniques, transformer-based models have shown promising results in analysing sequential data in natural language processing which may be extended to protein amino acid sequence. In this paper, a novel technique for efficiently and accurately predicting PPIs utilizing transformer models is proposed. In order to predict the capability of interaction, our proposed model quickly evaluates a pair of provided FASTA protein sequences as input leveraging decoder-only transformer model. This enables our model to predict PPIs with previously unprecedented accuracy. The findings reveal how this technique may be used for large-scale prediction tasks, allowing researchers to quickly and precisely discover possible new protein-drugs and targets. This finding has important ramifications as it may help find novel therapeutic protein-based medications by identifying proteins that interact with target proteins associated with diseases.