Accurate single-sequence prediction of solvent accessible surface area using local and global features

Area of Science:

• Computational biology
• Protein structure prediction
• Bioinformatics

Background:

• Accurate prediction of protein Accessible Surface Area (ASA) is crucial for understanding protein structure and function.
• Traditional methods often rely on computationally intensive multiple sequence alignments to generate residue mutation profiles.

Purpose of the Study:

• To develop a novel, efficient, and accurate method for predicting protein ASA.
• To explore the utility of sequential window information and global features as alternatives to multiple sequence alignments.

Main Methods:

• A General Neural Network (GENN) architecture was employed for ASA prediction.
• The model utilizes sequential window information and global features (single-residue and two-residue compositions) instead of mutation profiles.
• The predictor, named ASAquick, was evaluated on globular proteins.

Main Results:

• ASAquick demonstrates high efficiency compared to sequence alignment-based predictors.
• The predictor achieves comparable accuracy to existing methods.
• Global input features significantly contribute to the achieved accuracy.
• The method performs consistently on both 'easy' and 'hard' cases, indicating generalizability.

Conclusions:

• ASAquick offers an efficient and accurate alternative for ASA prediction.
• The approach's generalizability suggests potential applications in de-novo protein structure prediction.
• The developed GENN model and ASAquick predictor are publicly available.