A biofoundry automates the design-build-test-learn cycle of synthetic biology so that machine-learning models can steer each new round of genetic designs toward better performance, based directly on the last round's test data.
fitness(t+1) = fitness(t) + throughput*mlFeedback*(learnedGradient)
- DBTL cycle nodes — the Design, Build, Test and Learn stages a genetic construct passes through each iteration.
- Automation connectivity — how tightly robotic liquid handlers and sequencers link each DBTL stage without manual handoff.
- Cycle throughput — how many construct variants can be pushed through a full DBTL loop per unit time.
- ML feedback strength — how effectively the Learn stage's model steers the next Design round toward better variants.
Automated biofoundries (Ginkgo Bioworks, the Berkeley/JBEI foundry) can run thousands of DBTL cycles per week, turning strain engineering from an artisanal, one-off process into something closer to industrial R&D at scale.