OrthoML2GO: homology-based protein function prediction using orthogroups and machine learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The increasing volume of sequencing data presents challenges for protein sequence functional annotation.
• Traditional homology-based methods struggle with distant homologs, limiting accurate protein function determination.

Purpose of the Study:

• To introduce OrthoML2GO, a novel method for enhanced protein function prediction.
• To improve the accuracy and efficiency of protein annotation, particularly for large and heterogeneous datasets.

Main Methods:

• Integration of homology searches (USEARCH), orthogroup analysis (OrthoDB v12.0), and machine learning (gradient boosting).
• Sequential application of k-nearest neighbors (KNN), orthogroup annotation, and machine learning verification for GO term refinement.
• Comparative analysis of OrthoML2GO against Blast2GO and PANNZER2 using diverse organismal samples.

Main Results:

• OrthoML2GO demonstrates comparable or superior performance to existing methods in protein function prediction accuracy.
• The method shows particular strength in predicting functions for large and evolutionarily diverse protein datasets.
• Combining closest homolog information, orthogroups, and machine learning significantly enhances prediction performance.

Conclusions:

• OrthoML2GO offers a high-performance solution for large-scale automatic protein annotation.
• Future development can focus on optimizing machine learning models and integrating additional structural/functional data for improved accuracy and versatility.