Prediction of Enzyme Function Based on Three Parallel Deep CNN and Amino Acid Mutation

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• The rapid expansion of protein databases necessitates advanced computational methods for enzyme function prediction.
• Traditional approaches relying solely on protein sequence or structure are insufficient for understanding novel enzyme activities.
• Accurate enzyme function prediction is vital for various applications in biochemistry and biotechnology.

Purpose of the Study:

• To develop a novel computational model for predicting enzymatic function by integrating diverse biological information.
• To enhance the accuracy of enzyme function prediction beyond existing sequence-only or structure-only methods.
• To leverage deep learning architectures for simultaneous analysis of sequence and structural features.

Main Methods:

• Extraction of mutation information from amino acid sequences using position scoring matrices.
• Representation of protein structural information via amino acid distances and angles.
• Utilizing histograms for feature extraction from sequence and structural data.
• Employing a parallel Deep Convolutional Neural Network (DCNN) model with fused outputs from two architectures.

Main Results:

• The proposed model achieved 92.34% correct classification on a large dataset of 43,843 enzymes from the Protein Data Bank (PDB).
• Integration of sequence and structural information significantly improved prediction accuracy compared to previous methods.
• Demonstrated the efficacy of DCNNs in learning complex patterns from multi-faceted biological data.

Conclusions:

• The developed computational model effectively predicts enzyme function by integrating sequence and structural data.
• This approach offers a significant advancement in understanding enzyme roles in chemical reactions.
• The findings highlight the potential of deep learning in bioinformatics for biological discovery.