Encoding Genetic Circuits with DNA Barcodes Paves the Way for Machine Learning-Assisted Metabolite Biosensor Response Curve Profiling in Yeast

Genetically
encoded biosensors are valuable tools used in the precise
engineering of metabolism. Although a large number of biosensors have
been developed, the fine-tuning of their dose–response curves,
which promotes the applications of biosensors in various scenarios,
still remains challenging. To address this issue, we leverage a DNA
trackable assembly method and fluorescence-activated cell sorting
coupled with next-generation sequencing (FACS-seq) technology to set
up a novel workflow for construction and comprehensive characterization
of thousands of biosensors in a massively parallel manner. An FapR-fapO-based malonyl-CoA biosensor was used as proof of concept
to construct a trackable combinatorial library, containing 5184 combinations
with 6 levels of transcription factor dosage, 4 different operator
positions, and 216 possible upstream enhancer sequence (UAS) designs.
By applying the FACS-seq technique, the response curves of 2632 biosensors
out of 5184 combinations were successfully characterized to provide
large-scale genotype–phenotype association data of the designed
biosensors. Finally, machine-learning algorithms were applied to predict
the genotype–phenotype relationships of the uncharacterized
combinations to generate a panoramic scanning map of the combinatorial
space. With the assistance of our novel workflow, a malonyl-CoA biosensor
with the largest dynamic response range was successfully obtained.
Moreover, feature importance analysis revealed that the recognition
sequence insertion scheme and the choice of UAS have a significant
impact on the dynamic range. Taken together, our pipeline provides
a platform for the design, tuning, and profiling of biosensor response
curves and shows great potential to facilitate the rational design
of genetic circuits.