Neural Network and Random Forest Models in Protein Function Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The rapid increase in protein sequencing necessitates automated methods for function prediction.
• Accurate assignment of Gene Ontology (GO) terms is vital for understanding protein roles.

Purpose of the Study:

• To develop and evaluate an ensemble machine learning system for automated protein function prediction.
• To improve upon existing methods by combining diverse modeling approaches.

Main Methods:

• An ensemble system was created by merging predictions from Random Forest (RF) and Neural Network (NN) classifiers.
• Features were derived from BLAST alignments, taxonomy, and protein signatures.
• A novel NN model directly processed amino acid sequences using a convolutional layer.

Main Results:

• The ensemble model achieved competitive performance in the CAFA3 evaluation, ranking among the top 10 out of over 100 systems.
• The system demonstrated robust protein function prediction capabilities based on experimental verification.
• Further improvements were made to the CAFA3-submitted system.

Conclusions:

• Ensemble machine learning approaches effectively predict protein function.
• The developed system provides a valuable tool for bioinformatics research.
• The open-source release facilitates further development and application in protein science.