Aptamer- Protein Interaction Prediction using Transformer

Abstract

Traditional drug discovery has focused on antibody which costs a lot of time and efforts to produce it. To speed up drug discovery processes, various new biomaterial has been developed such as aptamers that are short single-strand oligonucleotide with three-dimensional structure. While aptamers show similar binding affinity to antibody, it is cheaper and faster than antibody. Systematic evolution of ligands by exponential enrichment (SELEX) is an in vitro experimental method to discover aptamers that bind a target protein. It takes several months to perform the full round of the SELEX experiment. To reduce time and cost for SELEX, there are several studies that find aptamers in silico, but most focus on the analysis of SELEX experiment results. There are some studies that apply machine learning to predict the interaction of aptamers and a target protein, but they feed only primary structure of aptamers and proteins into their machine model while apatmers and proteins are in three-dimension. This causes information loss. In this study, we propose a new machine learning model based on Transformer, which inputs aptamers and proteins in secondary structure. We validate our model using benchmark datasets and compare it with four existing methods. Our model outperforms in this evaluation. We believe that our model can improve the efficiency of SELEX experiments.