Pathway Design through Machine Learning

Leverage machine learning and probabilistic modeling techniques to guide synthetic biology in a systematic fashion

Given the growing amount of available synthetic biology parts (promoters, ribosome binding sites, enzymes, terminators, etc.), it is becoming increasingly complicated to choose the optimals for a given pathway. The traditional approach of creating libraries with all possible combinations is becoming less and less practical as the number of possible parts grows. 

Our machine learning algorithms (ART: the Automated Recommendation Tool) bridge Learn and Design, and allow us to test only a small fraction of all possible combinations, infer the results for all others, and recommend optimal combinations. We have shown this capability with the recommendation of optimal promoters, but the same approach can be used with other parts: ribosome binding sites of different strengths, enzymes of different kinetic characteristics, terminators, etc. 

This capability is ideally suited for combination with ABF’s high-throughput DNA construction and high-throughput strain characterization capabilities

Labs:
  • Lawrence Berkeley National Laboratory