Out-of-Equilibrium Selection Pressure Enhances Inference from Protein Sequence Data

Area of Science:

• * Computational biology
• * Molecular evolution
• * Bioinformatics

Background:

• * Homologous proteins share similar structures and functions, leading to coevolutionary correlations in their sequences.
• * Coevolutionary methods, including Potts models and AlphaFold, leverage these correlations to predict protein structure and function.
• * Understanding factors influencing inference accuracy is crucial for advancing protein science.

Purpose of the Study:

• * To investigate the impact of fluctuating selection strength and new selection pressures on coevolution-based protein inference.
• * To determine whether natural variations in selection dynamics enhance or impede the accuracy of protein sequence analysis.
• * To extend findings to realistic datasets and the prediction of protein interaction partners.

Main Methods:

• * Employed a minimal theoretical model to simulate protein sequence evolution under varying selection pressures.
• * Analyzed the effect of fluctuating selection strength and the introduction of new selective constraints.
• * Validated model predictions using realistic synthetic protein sequence data.

Main Results:

• * Demonstrated that fluctuating selection strength significantly improves the inference of structural contacts from protein sequences.
• * Showed that the emergence of new selection pressures also enhances coevolution-based inference accuracy.
• * Confirmed these improvements extend to predicting protein interaction partners.

Conclusions:

• * Out-of-equilibrium noise, driven by natural selection variations, is beneficial for coevolution-based protein inference.
• * Contrary to expectations, dynamic selection pressures enhance, rather than hinder, the success of predicting protein structure and function.
• * Findings provide new insights into the robustness and potential of sequence-based protein analysis methods.