Rapid and Interpretable Protein Contact Map Prediction Using a Pattern-Matching Strategy

Abstract

Protein sequence determines the structure, function, and dynamics of a protein. In recent years, enormous progress has been made in translating sequence information into structural information using machine learning approaches. However, because of the underlying methodology, it is an immense computational challenge to extract this information from the ever-increasing number of sequences. In the present study, we show that it is possible to create two-dimensional contact maps from sequences, for which only a few exemplary structures are available on a laptop without the need for GPUs or high-performance computing clusters. This is achieved by using a pattern matching approach. The resulting contact maps largely reflect the interactions in the three-dimensional structures. The validity of our method was tested on the 25 protein domains, with abundant structural data, achieving correlations of 0.73-0.94 between predicted and experimental contact maps. To demonstrate broader applicability, we further validated our approach on 7,599 poorly annotated sequences using homologous structural templates, achieving a mean F1-score of 0.609 {+/-} 0.095 and mean accuracy of 0.954 {+/-} 0.036 when compared against high-confidence AlphaFold structures. These results demonstrate that our pattern matching approach maintains robust performance even when relying on a small number of structural templates.