Location: Genomics and Bioinformatics Research
Project Number: 6066-21310-006-037-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Aug 15, 2024
End Date: Aug 14, 2027
Objective:
The use of probiotics and inoculants shows promise for controlling plant disease, and improving animal health and feed efficiency the field is currently hampered because there is no way to select the optimal combination of microbes in a probiotic or inoculant formulation. This project develops a microfluidic system for analyzing up to 100,000 microbial consortia in parallel, and the statistical tools to use this data as a training set to optimize consortia with generative artificial intelligence methods.
Approach:
The cooperator will create semipermeable microfluidic capsules containing a fluorescently labeled target pathogen and a random mixture of microorganisms that are candidates for inclusion in probiotic microbial consortia. The capsules will be sorted by flow cytometry to retain capsules containing the pathogen. The capsules will be incubated in culture media, then washed lysed and Split pool-barcoding will be used to uniquely label the leach community and amplicon sequence the internal transcribed spacer region one using Illumina. That data will provide proportional data on the pathogen and the members of the inoculant consortia for 50,000 consortia. Those training data will be used to train a regression model predicting the abundance of the target pathogen in the community based on the final concentrations of the consortia members. The regression model will be combined with a disentangled variational autoencoder and an adaptive sampling algorithm to create a generative optimization system based on conditioning by adaptive sampling. The end result will be a system that predicts consortia more suppressive than any in the training dataset. Those predictions will be validated in the lab by merging microfluidics droplets to generate specific communities.
