Computer environment supporting biosystem design guided by machine learning

BioSynthetic Machines, Inc. (BSMI) is a Chicago, IL-based startup company launched in late 2021 with a goal to revolutionize design and engineering of the microorganisms for production of chemicals. We utilize the innovative machine learning (ML) and experimental methods of synthetic biology developed by the founders at Argonne National Laboratory and exclusively licensed to the company. Our proof-of-principle de novo engineering of threonine-producing strains serves as a validation of our technology. In a pilot study, 600 strains carrying pairwise combinations of modifications of 20 genes selected via metabolic pathway analysis and inserted in in two bacterial hosts, were subjected to two rounds of ML-driven engineering cycles in which we analyzed 107- variant virtual space of combinations. The best threonine producers resulting from this pilot were 2.5X times more productive (in laboratory conditions) than the industrial control strain. Although the company is very young, BSMI has an exceptionally strong position in synthetic biology, with the key personnel and advisory team having an extensive list of publications and established track record in ML, computational biology, database construction, metabolic modelling, microbiology, and industrial fermentation. Our SBIR project team has extensive expertise in company-wide management for construction of industrial microorganisms used by large-scale commercial processes. Development of genome-wide annotation techniques revolutionized the world of functional interpretation of genomes and were implemented in the widely accepted SEED and RAST platforms in which BSMI team members played important roles. Our aspiration is to use proprietary AI, computational, and experimental approaches to synthetic biology to change organism engineering for microbiological production of chemicals. Our business goal is to deliver value-added and innovative strain development services and groundbreaking new products that address real biomanufacturing market needs. In this SBIR project we plan to develop a novel and yet unproven hybrid ML approach to combinatorial gene target selection that would integrate knowledge-based ML model design with agnostic iterations of target prediction/validation. This project aims to add hypotheses- based-on-a-prior-knowledge element to ML in the construction of new superior bio producents. Engineering of organisms for biotechnological purposes provides rich data for future analysis which has to be stored, curated and integrated in the strain design process; for this, we are developing a sophisticated and user-friendly tool for a storage, integration, and alignment of multiple data types, and it is to be populated during this SBIR project. (1) a validated hybrid-ML approach for strain engineering applicable to a broad range of biomanufacturing projects; (2) a specific set of proprietary software tools supporting this approach, and (3) an improved threonine-producing strain for commercial use. Our innovative solution to the complexities of organism construction is a combination of hypothesis-driven MM-based data refinement and Accelerated “Agnostic” Strain Engineering (AASE), a modification of a generic Design-Build-Test-Learn (DBTL) approach which relies on ML and combinatorial gene engineering. Such hybrid approach, assisted by a prior knowledge, where it is available, is universal in nature and can dramatically expand the list of chemicals produced in biomanufacturing by eliminating its biggest bottleneck - the slow, expensive, and poorly predictable process of strain development.