Abstract: 

Living cells naturally perform information processing using genetic circuits — networks of DNA-encoded genes that dynamically sense and respond to their environment. The ability to engineer new genetic circuits has broad applications in biotechnology, personalized medicine and human health. Biologically informed, predictive, and robust genetic circuit design tools are necessary to realize the next generation of intelligent treatments.

Despite progress over the past two decades, genetic circuit design remains imprecise. Computer-aided design tools have been coupled with genetic part libraries to design complex genetic circuits. However, many circuits fail to function as predicted due to hidden genetic interactions, contextual effects at the DNA-level, and other subtle biophysical phenomena.

Here, we discuss a radically new approach for designing and inferring performance of genetic circuits. We use machine learning to fit complex biophysical models that map to semantic structures to enable more accurate forward inference of circuit performance. Methods adapted from Bayesian inference, sequence-to-sequence models and deep generative models are applied to realize and introspect biophysical systems.

Bio: 

Joe Isaacson is the vice president of engineering at Asimov, a biotechnology startup that engineers living cells to produce next generation therapeutics. Asimov develops machine learning methods to accelerate the R&D lifecycle of synthetic biology. Previous to Asimov, Joe lead machine learning teams at Quora (a top 100 Alexa website) and URX (acquired by Pinterest) building products spanning recommendation systems, ad targeting and search relevancy. Joe has spent the past ten years contributing to research and products spanning forensics, computer vision, natural language understanding and biological design.