Discovering genotype-phenotype relationships with machine learning and the Visual Physiology Opsin Database (VPOD).

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Predicting phenotypes from genetic variation is crucial for bioengineering and global change biology.
• Opsin proteins, key to color vision, have their genotype-phenotype relationships extensively studied but data is fragmented.
• Understanding opsin genotype-phenotype links requires accessible, compiled data.

Purpose of the Study:

• To compile a comprehensive database of opsin genotypes and their λmax phenotypes.
• To develop and validate machine learning models for predicting opsin λmax from genetic sequence.
• To facilitate systematic analysis of genotype-phenotype relationships in opsins.

Main Methods:

• Compiled the Visual Physiology Opsin Database (VPOD) from 73 publications, including 864 opsin genotypes and their λmax phenotypes.
• Utilized deepBreaks and regression-based machine learning (ML) models.
• Analyzed VPOD data to train and test ML models for λmax prediction.

Main Results:

• The VPOD (version 1.0) contains 864 unique opsin genotypes and associated λmax phenotypes.
• Machine learning models reliably predict λmax, capturing nonadditive mutation effects.
• Identified functionally critical amino acid sites within opsin sequences.

Conclusions:

• ML enables reliable prediction of opsin function from gene sequences alone.
• This facilitates exploration of molecular evolution, ecological niche connections, and de novo protein design.
• The database and predictive models provide a foundation for future research on quantifiable phenotypes.