Relating amino acid sequence to phenotype: analysis of peptide-binding data

Area of Science:

• Genomics
• Immunology
• Bioinformatics

Background:

• Relating genotype (amino acid sequences) to phenotypes (outcomes) presents data analytic challenges.
• Predicting peptide binding to major histocompatibility complex (MHC) class I molecules is crucial but complex.
• Simple prediction rules based on specific amino acid positions are often inadequate due to sequence-wide interactions.

Purpose of the Study:

• To address data analytic concerns in genotype-phenotype association studies, specifically for peptide-MHC binding.
• To develop and demonstrate effective methods for modeling complex interactions within amino acid sequences.
• To investigate whether observed associations are attributable to specific amino acid properties or sequence interactions.

Main Methods:

• Utilizing classification trees to handle multilevel, unordered categorical covariates (amino acids) and their interactions.
• Employing bump-hunting methods to identify significant associations and interactions.
• Comparing sequence-based analysis with methods relying on continuous amino acid properties.

Main Results:

• Classification trees and bump-hunting effectively model between-position interactions in amino acid sequences.
• These methods overcome limitations of standard regression techniques when dealing with numerous categorical covariates and interactions.
• The study assesses the contribution of amino acid properties versus sequence interactions to binding outcomes.

Conclusions:

• Advanced statistical methods like classification trees and bump-hunting are essential for accurate genotype-phenotype association, particularly in complex biological sequences.
• These approaches provide a more nuanced understanding of peptide-MHC binding than simpler models.
• The findings highlight the importance of considering all sequence positions and their interactions for robust biological predictions.