Impact of phylogeny on structural contact inference from protein sequence data

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein structure prediction from sequence data is crucial for understanding protein function.
• Existing methods rely on correlations in amino acid usage within homologous protein sequences.
• Phylogenetic correlations, arising from common ancestry, can interfere with accurate contact inference.

Purpose of the Study:

• To investigate the impact of phylogenetic correlations on protein contact inference methods.
• To compare the resilience of local versus global inference methods to phylogenetic effects.
• To understand how evolutionary factors influence the accuracy of contact prediction.

Main Methods:

• Generation of controlled synthetic data from a minimal protein sequence model.
• Tuning the influence of contacts and phylogeny in the synthetic data.
• Evaluation of local (covariance, mutual information) and global (Potts models) inference methods.
• Analysis of natural and realistic synthetic protein sequence data.

Main Results:

• Global inference methods, particularly Potts models, demonstrate greater robustness against phylogenetic correlations compared to local methods.
• This resilience holds true regardless of whether phylogenetic corrections are applied.
• Early-mutating sites in the phylogeny were identified as sources of false positive contacts.

Conclusions:

• Global inference methods offer superior performance in protein contact prediction due to their inherent resistance to phylogenetic noise.
• Phylogeny significantly impacts contact prediction accuracy, highlighting the importance of considering evolutionary history.
• Understanding the interplay between biological data structure and inference algorithms is key for advancing protein science.