Guiding Mineralization Co-Culture Discovery Using Bayesian Optimization.

Area of Science:

• Microbiology
• Environmental Science
• Data Science

Background:

• Scientific discovery often requires testing numerous combinations of biological or chemical entities.
• Experimental determination of optimal combinations is frequently time-consuming and resource-intensive.
• Rational, data-driven approaches are needed to enhance experimental efficiency.

Purpose of the Study:

• To demonstrate the utility of machine learning for in silico selection of bacterial co-cultures for bioaugmentation.
• To develop a data-driven experimental design for identifying high-performing co-cultures.
• To reduce the experimental burden in discovering effective microbial consortia for pollutant removal.

Main Methods:

• Utilized Gaussian process regression to model bacterial co-culture performance based on single-strain parameters.
• Employed Bayesian optimization to suggest promising co-culture combinations for pollutant degradation.
• Applied the methods to a dataset of bacterial pairs co-cultured with MSH1 for mineralization in a drinking water treatment context.

Main Results:

• Achieved accurate prediction of co-culture mineralization parameters using the Gaussian process regression model.
• Successfully identified effective bacterial co-cultures for pollutant removal via Bayesian optimization.
• Demonstrated good performance of the machine learning approach despite a limited dataset.

Conclusions:

• Machine learning, specifically Bayesian optimization, offers a powerful and efficient method for selecting bacterial co-cultures in bioremediation.
• This data-driven experimental design significantly reduces the effort required for identifying optimal microbial consortia.
• The approach holds promise for various applications requiring targeted microbial screening and optimization.