Reciprocal Best Structure Hits: Using AlphaFold models to discover distant homologues

1 Abstract The conventional methods to detect homologous protein pairs use the comparison of protein sequences. But the sequences of two homologous proteins may diverge significantly and consequently may be undetectable by standard approaches. The release of the AlphaFold 2.0 software enables the prediction of highly accurate protein structures and opens many opportunities to advance our understanding of protein functions, including the detection of homologous protein structure pairs. In this proof-of-concept work, we search for the closest homologous protein pairs using the structure models of five model organisms from the AlphaFold database. We compare the results with homologous protein pairs detected by their sequence similarity and show that the structural matching approach finds a similar set of results. Additionally, we detect potential novel homologues solely with the structural matching approach, which can help to understand the function of uncharacterised proteins and make previously overlooked connections between well-characterised proteins. We also observe limitations of our implementation of the structure based approach, particularly when handling highly disordered proteins or short protein structures. Our work shows that high accuracy protein structure models can be used to discover homologous protein pairs, and we expose areas for improvement of this structural matching approach.