A discriminative method for family-based protein remote homology detection that combines inductive logic programming...

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• Remote homology detection is a significant computational challenge.
• Traditional methods using full sequences or MSAs struggle with proteins in the 'twilight zone' where only motifs are conserved.
• Identifying conserved sequence segments (motifs) is crucial for understanding protein function.

Purpose of the Study:

• To develop a novel computational method for remote homology detection.
• To represent physico-chemical properties and conserved positions within protein sequences and MSAs.
• To leverage Inductive Logic Programming (ILP) for motif discovery and model training.

Main Methods:

• Introduced a novel logical representation for protein sequence properties.
• Utilized Inductive Logic Programming (ILP) to find frequent conserved patterns (motifs).
• Trained propositional models, including decision trees and Support Vector Machines (SVMs), using identified motifs.

Main Results:

• The methodology was evaluated using the SCOP database for protein recognition within superfamilies.
• The SVM-based approach demonstrated significantly improved performance compared to some state-of-the-art methods.
• The method generated comprehensible logical rules, aiding in the understanding of protein function determinants.

Conclusions:

• Selecting frequent patterns is an effective strategy for remote homology detection.
• A first-order logical representation of homologous properties is key.
• ILP-identified frequent patterns summarize essential protein function features.