Machine learning can distinguish orphans that have resulted from sequence divergence beyond recognition.

Area of Science:

• Genomics
• Bioinformatics
• Evolutionary Biology

Background:

• Species-specific orphan genes contribute to unique traits but are hard to identify due to extreme sequence divergence.
• Traditional methods discard non-statistically significant similarity hits, potentially missing divergent homologous genes.

Purpose of the Study:

• To develop a machine learning approach for identifying highly divergent orphan genes.
• To investigate the characteristics of these potentially divergent orphan genes.

Main Methods:

• Simulated diverged orphan protein sequences under various parameters.
• Trained machine learning classifiers on features from similarity search outputs, using reversed sequences as a negative control.
• Applied trained classifiers to a set of real orphan genes.

Main Results:

• Machine learning models achieved up to ~90% accuracy in identifying moderately diverged simulated orphans and ~70% for extremely diverged ones.
• Approximately 30% of real orphan genes were predicted as divergent.
• Predicted divergent orphans were found to be shorter and more disordered compared to other orphans.

Conclusions:

• Subtle patterns in similarity search outputs can reveal highly diverged homologous genes.
• This method enhances the identification of genetic novelty and aids in understanding the evolution of species-specific traits.
• The developed models and data are publicly available for further research.