A continuous epistasis model for predicting growth rate given combinatorial variation in gene expression and...

Area of Science:

• Microbiology
• Systems Biology
• Computational Biology

Background:

• Predicting bacterial growth rate phenotypes is challenging due to complex gene interactions (epistasis) and environmental factors.
• Understanding how gene expression variations influence growth is crucial for various biological and biotechnological applications.

Purpose of the Study:

• To develop and validate an interpretable machine learning approach for mapping gene expression-growth rate landscapes.
• To quantify epistatic interactions in bacterial gene expression under diverse environmental conditions.

Main Methods:

• Utilized mismatch CRISPR interference (CRISPRi) to create over 8,000 titrated gene expression changes in E. coli.
• Applied an interpretable machine learning model integrating sparsely sampled experimental data across up to 22 distinct environments.
• Explored pairwise and triple gene perturbations to assess epistatic effects on growth rate.

Main Results:

• A pairwise interaction model, previously used for drug interactions, effectively described the gene expression-growth rate data.
• The model provided interpretable parameters reflecting pathway architecture and successfully predicted the combined effects of up to four gene perturbations.
• Demonstrated the model's generalization capability when trained on pairwise perturbation data alone.

Conclusions:

• The developed machine learning approach accurately models bacterial growth rate phenotypes influenced by gene expression and environmental context.
• This method offers a powerful tool for understanding bacterial gene expression constraints and optimizing growth conditions.
• The approach has broad applicability in areas such as pharmacogenomics and synthetic biology.