Protein Contact Map Stability Affect Sequence Co-Evolution

Several computational methods utilize co-evolution information from Multiple-Sequence-Alignments (MSA) of a protein family to predict the physical contact map of a target protein from the family, and subsequently to improve protein 3D structure prediction.

The methods rely on the hypothesis that amino-acids that are in physical contact show compensatory mutations that preserve the structure and function of the protein, while mutations that disrupt the local or global structure are selected against and are observed at lower frequencies.

However, many protein families show structural heterogeneity. In particular, some of the physical contacts are evolutionary unstable and may be present only in part of the alignment. Current methods utilize the entire alignment columns, thus ignoring contact instability which may lead to reduced signals or false positives.

Here, we studied the changes in co-evolution of amino-acid pairs across the alignment. We developed a method for detection of changes in pairwise co-evolution in a MSA, by scanning sequences in the family based on their sequence distance from the target protein, and identifying the sub-alignments which show the strongest co-evolutionary signals.

We used this method to filter alignments and showed that using the filtered list improves prediction accuracy for unstable contacts in the target protein.

We then utilized this information to develop a novel method for structure-assisted contact prediction, where we used a deep learning approach to predict the contact map for a target protein, utilizing both the co-evolutionary signal from the alignment, and structural information from known structures within the same family