LARGE-SCALE MULTIPLE INFERENCE OF COLLECTIVE DEPENDENCE WITH APPLICATIONS TO PROTEIN FUNCTION

Area of Science:

• Computational Biology
• Information Theory
• Statistical Inference

Background:

• Analyzing higher-order dependencies (k ≥ 3) among random variables is crucial but challenging.
• Protein coevolution, involving multivariate categorical features, requires advanced methods beyond pairwise analysis.

Purpose of the Study:

• To introduce and validate a novel information-theoretic measure, collective dependence, for quantifying higher-order variable interactions.
• To develop a robust statistical inference procedure for detecting significant k-collective dependence in large datasets.
• To apply the method to protein sequence data for uncovering novel coevolving residue insights.

Main Methods:

• Defined collective dependence as a symmetrized measure generalizing mutual information for k ≥ 3 variables.
• Developed a Classification-Assisted Large scaLe inference procedure (CALDECO) for detecting significant k-collective dependence with controlled false discovery rate.
• Validated the method through simulations and applied it to protein sequence alignment data.

Main Results:

• Collective dependence is easily estimated and facilitates dependence testing for k ≥ 3 variables.
• CALDEDECO successfully identified significant higher-order co-dependencies in simulated data.
• Novel functional triplets of amino acid residues were identified in elongation factor P and zinc knuckle protein families.

Conclusions:

• Collective dependence provides valuable insights into protein coevolution, surpassing traditional pairwise measures.
• The developed CALDECO procedure offers a powerful tool for large-scale inference of complex variable interactions.
• Identified coevolving residue triplets offer new avenues for investigating protein function and mechanisms.