Improvement of the GenTHREADER method for genomic fold recognition

Area of Science:

• Computational biology
• Bioinformatics
• Structural bioinformatics

Background:

• The GenTHREADER method, a fully automatic fold recognition tool, was enhanced to improve genome annotation.
• The previous version utilized a simple neural network combining sequence alignment scores, length information, and energy potentials.
• The improved version integrates PSI-BLAST searches, PSIPRED secondary structure predictions, and bi-directional scoring.

Purpose of the Study:

• To enhance the accuracy and reliability of the GenTHREADER fold recognition method.
• To improve genome annotation through more effective detection of remote protein homologues.
• To benchmark the performance of the improved GenTHREADER against its previous version.

Main Methods:

• Incorporation of PSI-BLAST searches, initiated with FSSP structural alignment profiles.
• Utilization of PSIPRED-predicted secondary structure and bi-directional scoring for alignment score calculation.
• Expansion of the multi-layer feed-forward neural network to include secondary structure element alignment (SSEA) score and training to learn FSSP Z-score.

Main Results:

• Increased detection of remote homologues with a low error rate, indicating higher score reliability.
• Up to five times more true positives detected per query at low false positive rates.
• Doubled Total MaxSub score at low false positive rates, signifying improved model quality.

Conclusions:

• The enhanced GenTHREADER method significantly improves remote homology detection and protein model quality.
• The improvements contribute to more reliable genome annotation through advanced computational approaches.
• The updated tool offers a more powerful and accurate solution for protein structure prediction and analysis.