GCmapCrys: Integrating graph attention network with predicted contact map for multi-stage protein crystallization...

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• X-ray crystallography is a primary method for determining protein structures at the atomic level.
• The success of X-ray crystallography is often limited by the difficulty in crystallizing proteins.
• Accurate prediction of protein crystallization propensity is essential for optimizing experimental design and increasing success rates.

Purpose of the Study:

• To develop a novel deep learning pipeline, GCmapCrys, for enhanced prediction of protein crystallization propensity.
• To integrate graph attention networks with predicted protein contact maps for improved predictive performance.

Main Methods:

• Development of the GCmapCrys deep learning pipeline.
• Integration of graph attention networks (GAT) with predicted protein contact maps.
• Utilizing four novel sequence-based features to complement predictions.

Main Results:

• GCmapCrys demonstrated a significant improvement in predicting protein crystallization propensity.
• The pipeline achieved an average increase of 37.0% in Matthew's correlation coefficient compared to existing state-of-the-art predictors.
• Analysis highlighted the effectiveness of combining GAT with predicted contact maps for capturing residue interaction patterns relevant to crystallization.

Conclusions:

• GCmapCrys offers a powerful new tool for predicting protein crystallization propensity.
• The integration of graph attention networks and contact maps provides a more accurate and efficient approach.
• The developed method can guide experimental efforts in X-ray crystallography, potentially accelerating protein structure determination.