Discovery of a phenazine-thiol conjugase from sparse data using genome-informed machine learning.

Area of Science:

• Biochemistry
• Genomics
• Machine Learning

Background:

• Machine learning (ML) typically requires large datasets, posing a challenge for biological discovery where data is often sparse.
• Identifying enzymes for understudied substrates is hindered by limited training data.

Purpose of the Study:

• To develop a method for enzyme discovery using limited biological data by integrating genome-informed data augmentation with protein machine learning.
• To identify novel enzymes involved in modifying phenazine natural products.

Main Methods:

• Utilized genome-informed data augmentation combined with contrastive learning in protein language space.
• Trained models on a small dataset of 14 known phenazine-modifying sequences.
• Applied the framework to identify phenazine-interacting proteins.

Main Results:

• Discovered Phenazine-Thiol Conjugase (PTC), the first known enzyme catalyzing phenazine thioconjugation.
• PTC binds to both phenazine and glutathione, facilitating glutathione-dependent phenazine modification.
• PTC's function is essential for certain cellular processes, yet its absence does not cause a significant fitness disadvantage, highlighting enzymes missed by phenotype-based screens.

Conclusions:

• Coupling comparative genomics with protein ML effectively addresses "small data" challenges in enzyme discovery.
• This approach enables the identification of enzymes with crucial but potentially subtle biological roles.
• The discovery of PTC expands the known repertoire of enzymatic reactions in natural product modification.