Protein engineering aims to identify variants with properties useful for biological, industrial, or medical purposes.
1 As a powerful protein engineering tool, continuous evolution enables gene diversification and screening in an iterative manner without human intervention. However,
in vivo continuous evolution can rarely be applied to proteins whose phenotypes are unrelated to cell growth. It is also challenging to adapt existing continuous evolution methods for new target proteins.
2 To address these limitations, we developed a flexible and versatile
in vitro continuous evolution workflow by integrating synthetic biology, robotics automation, and machine learning. While automation can expedite the directed evolution workflow and generate data at a capacity beyond manual work, it alone is not enough to perform directed evolution continuously. Thus, we use a machine learning model to predict and infer beneficial mutations and guide the design of continuous evolution. As a proof-of-concept, we used phytase as a case study to develop the iterative Design-Build-Test-Learn (DBTL) cycle in a biofoundry named BioAutomata.
3 The cycle was initiated with an informed small library designed by EVmutation
4 with the size of 96. We automated the entire workflow of constructing the variant library using site-directed mutagenesis (SDM) using two-step PCR. The SDM accuracy reached as high as 90%, with automated primer optimization. The subsequent experimental steps including transformation, colony picking, protein expression and assay measurement were all fully automated using biofoundry without the need of human intervention. The collected data was then used to train a machine learning model under the low-N framework.
5 The beneficial mutants predicted by the machine learning model were constructed for subsequent cycles. Such DBTL cycle took 4 days and can be carried out iteratively, with new data being generated and fed to the machine learning model in each cycle and leading to increased prediction accuracy. This
in vitro continuous evolution workflow should be generally applicable to any target proteins even without prior knowledge or high-throughput screening methods.
(1) Yang, K. K.; Wu, Z.; Arnold, F. H. Machine-Learning-Guided Directed Evolution for Protein Engineering. Nat. Methods 2019, 16 (8), 687–694.
(2) Wang, Y.; Xue, P.; Cao, M.; Yu, T.; Lane, S. T.; Zhao, H. Directed Evolution: Methodologies and Applications. Chem. Rev. 2021, 121 (20), 12384–12444.
(3) HamediRad, M.; Chao, R.; Weisberg, S.; Lian, J.; Sinha, S.; Zhao, H. Towards a Fully Automated Algorithm Driven Platform for Biosystems Design. Nat. Commun. 2019, 10 (1), 5150.
(4) Hopf, T. A.; Ingraham, J. B.; Poelwijk, F. J.; Schärfe, C. P. I.; Springer, M.; Sander, C.; Marks, D. S. Mutation Effects Predicted from Sequence Co-Variation. Nat. Biotechnol. 2017, 35 (2), 128–135.
(5) Hsu, C.; Nisonoff, H.; Fannjiang, C.; Listgarten, J. Learning Protein Fitness Models from Evolutionary and Assay-Labeled Data. Nat. Biotechnol. 2022.
