Application of multiple sequence alignment profiles to improve protein secondary structure prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Bioinformatics

Background:

• Accurate protein secondary structure prediction is crucial for understanding protein function and structure.
• Neural network algorithms offer a promising approach for improving prediction accuracy.

Purpose of the Study:

• To investigate the impact of different multiple sequence alignment (MSA) profile types on neural network-based secondary structure prediction accuracy.
• To develop and evaluate a novel secondary structure prediction method, Jnet.

Main Methods:

• Training a neural network algorithm using various MSA profiles derived from the same sequence datasets.
• Comparing the performance of the new method (Jnet) against the existing PHD algorithm on a non-redundant protein dataset.
• Evaluating prediction accuracy using Q(3) and SOV2 metrics.

Main Results:

• The Jnet method achieved a maximum prediction accuracy of 76.4%, outperforming the PHD algorithm by 3.1% (Q(3)) and 4.4% (SOV2).
• High-confidence predictions (confidence value ≥ 5) reached 84% Q(3) accuracy, covering 68% of residues.
• Accurate predictions were also achieved for relative solvent accessibility (up to 86.6%).

Conclusions:

• Training neural networks with diverse MSA profile representations significantly enhances protein secondary structure prediction.
• The Jnet method represents a substantial improvement over existing algorithms and is available via the Jpred server and as a standalone program.