CLPred: a sequence-based protein crystallization predictor using BLSTM neural network.

Area of Science:

• Structural Biology
• Computational Biology
• Bioinformatics

Background:

• Protein structure determination is crucial for understanding biological functions.
• X-ray diffraction is the primary experimental method, but protein crystallization is challenging and costly.
• Current in silico methods for predicting crystallizability have limited predictive power.

Purpose of the Study:

• To develop an accurate deep learning model for predicting protein crystallizability using only sequence information.
• To improve upon existing computational approaches for identifying proteins amenable to crystallization.

Main Methods:

• A deep learning model named CLPred was developed.
• CLPred utilizes a bidirectional recurrent neural network with long short-term memory (BLSTM).
• The model captures long-range interaction patterns between amino acid k-mers.

Main Results:

• CLPred significantly outperforms existing deep learning and sequence-based predictors on three independent test sets.
• The BLSTM architecture effectively identifies non-local inter-peptide interactions relevant to crystal formation.
• Case studies on Sox transcription factors and Zika virus proteins validate CLPred's accuracy.

Conclusions:

• CLPred demonstrates high accuracy in predicting protein crystallization propensity.
• The model offers a valuable tool for selecting promising protein candidates for experimental structure determination.
• Future improvements are possible by integrating evolutionary, structural, and physicochemical features.