Multiple Profile Models Extract Features from Protein Sequence Data and Resolve Functional Diversity of Very...

Area of Science:

• Computational biology
• Bioinformatics
• Protein science

Background:

• Functional classification of proteins from sequences alone is a major challenge.
• Homologous protein sequences often exhibit diverse, unpredictable functions crucial for understanding evolution and biotechnology.

Purpose of the Study:

• To introduce ProfileView, a sequence-based computational method for functionally classifying homologous protein sequences.
• To improve the identification of protein functions, discover new functional groups, and understand protein functional diversity.

Main Methods:

• ProfileView utilizes multiple profile models incorporating evolutionary information.
• It employs a novel representation space for analyzing sequences with combined profile models.
• The method was validated on seven diverse protein classes involved in molecular interactions and enzymatic reactions.

Main Results:

• ProfileView successfully enriches known functional groups and discovers new ones.
• It accurately reflects literature-based functional subgroup organization and key residues.
• The method resolves ambiguous classifications and identifies molecular determinants of protein function.

Conclusions:

• ProfileView effectively classifies protein families, reconciles complex domain architectures, and outperforms existing methods like PANTHER, CUPP, eCAMI, and Restricted Boltzmann Machines.
• It offers potential for selecting sequences for experimental design and discovering novel biological functions.