Presentation Information
[2Biocat-12]Mining Microbial Dark Matter for Industrial Biotechnology
○Soichiro Tsuda1 (1. bitBiome Inc. (Japan))
Keywords:
Biomanufacturing,Microbial Dark Matter,Microbial Single-cell Genome Sequencing,Genome Foundation Model
The vast majority of microbial species on Earth remain uncultured, representing an enormous reservoir of genetic diversity largely inaccessible to conventional biotechnology. bitBiome has developed bit-MAP, a microfluidics-based single-cell whole-genome sequencing platform that recovers over 100-fold more non-redundant genes compared to shotgun metagenomic sequencing. This technology underpins bit-GEM, a proprietary database now exceeding 3 billion microbial gene sequences from diverse environments, such as soil, marine water, hot springs, and human microbiome, many of which are absent from public repositories.
Here we present several industrial applications enabled by our biomanufacturing platform: First, we developed a commercial transaminase screening kit comprising highly diverse enzyme candidates mined from bit-GEM, enabling broad-scope asymmetric amination reactions relevant to pharmaceutical and fine chemical synthesis. Second, using our AI-driven enzyme engineering pipeline (bit-QED), we identified and optimized novel PET hydrolase candidates from bit-GEM, achieving dramatically improved PET depolymerization activity through structure-informed directed evolution. This result clearly demonstrates how microbial dark matter can be harnessed for plastic biorecycling. Third, we have developed a high-throughput, massively parallel genome engineering platform that enables simultaneous introduction of diverse genomic modifications across large strain libraries. As a proof of concept, we applied this platform to Escherichia coli and successfully identified variants with up to 9-fold higher target protein production relative to the wild-type, through a single round of parallel genome editing and high-throughput screening. This capability can be potentially extended to non-model microorganisms, including CO2- and hydrogen-utilizing species, as a foundation for engineering next-generation biomanufacturing chassis that are not accessible by conventional strain development approaches.
Looking ahead, we are developing GEM-AI, a genome foundation model trained on bit-GEM's high-quality, full-length microbial gene sequences. Early benchmarking demonstrates that GEM-AI significantly outperforms existing foundation models in predicting functional enzyme variants, reflecting the superior data quality and diversity that single-cell genomics enables. Collectively, these results demonstrate that the integration of world-class microbial genomic data with generative AI represents a transformative approach to accelerating biocatalyst discovery, strain development, and ultimately the design of entirely novel biosynthetic pathways -- unlocking a new era of predictive and programmable biomanufacturing.
Here we present several industrial applications enabled by our biomanufacturing platform: First, we developed a commercial transaminase screening kit comprising highly diverse enzyme candidates mined from bit-GEM, enabling broad-scope asymmetric amination reactions relevant to pharmaceutical and fine chemical synthesis. Second, using our AI-driven enzyme engineering pipeline (bit-QED), we identified and optimized novel PET hydrolase candidates from bit-GEM, achieving dramatically improved PET depolymerization activity through structure-informed directed evolution. This result clearly demonstrates how microbial dark matter can be harnessed for plastic biorecycling. Third, we have developed a high-throughput, massively parallel genome engineering platform that enables simultaneous introduction of diverse genomic modifications across large strain libraries. As a proof of concept, we applied this platform to Escherichia coli and successfully identified variants with up to 9-fold higher target protein production relative to the wild-type, through a single round of parallel genome editing and high-throughput screening. This capability can be potentially extended to non-model microorganisms, including CO2- and hydrogen-utilizing species, as a foundation for engineering next-generation biomanufacturing chassis that are not accessible by conventional strain development approaches.
Looking ahead, we are developing GEM-AI, a genome foundation model trained on bit-GEM's high-quality, full-length microbial gene sequences. Early benchmarking demonstrates that GEM-AI significantly outperforms existing foundation models in predicting functional enzyme variants, reflecting the superior data quality and diversity that single-cell genomics enables. Collectively, these results demonstrate that the integration of world-class microbial genomic data with generative AI represents a transformative approach to accelerating biocatalyst discovery, strain development, and ultimately the design of entirely novel biosynthetic pathways -- unlocking a new era of predictive and programmable biomanufacturing.
Comment
To browse or post comments, you must log in.Log in