Identifying Candidates for Protein-Protein Interaction: A Focus on NKp46’s Ligands

Recent advances in protein-protein interaction (PPI) research have harnessed the power of artificial intelligence (AI) to enhance our understanding of protein behaviour. These approaches have become indispensable tools in the field of biology and medicine, enabling scientists to uncover hidden connections and predict novel interactions. The experimental processes to analyze and validate the interactions between proteins are usually expensive and time-consuming and with this work, we can reduce these costs by strategically filtering and computationally validating the possible proteins which might take part in the interactions at hand. Aiming at helping in broadening the repertoire of known interacting proteins, we present a method for the systematic screening of proteins that exhibit a high affinity for the interaction with a chosen protein. Specifically, building upon already known protein interactions, we exploit the self-explainability of the deep learning model DSCRIPT to search and find promising protein candidates for a determined PPI. We analyze and rank the candidates using various strategies, and then employ AlphaFold2 to validate the resulting interactions. Consequently, we compare our AI-driven methodology with traditional bioinformatics approaches commonly used to find potential protein candidates. Throughout the overall process, explanatory data is obtained, among which is an informative contact map that elucidates the potential interaction between a protein of the known interaction and the predicted proteins. As a case study, we apply our method to deepen our understanding of NKp46’s ligands repertoire, which is yet not fully uncovered.