Active learning guides automated discovery of DNA delivery via electroporation for non-model microbes

Delivery of recombinant DNA is foundational for understanding and engineering a target organism. Electroporation can be applied to any cell type, yet identification of a working protocol for new organisms remains bespoke and laborious because no systematic framework exists, and appropriate instrumentation is lacking. Here, we describe an automated high-throughput platform which uses active learning to discover electroporation protocols for non-model microbes. We first devised a 24-condition electroporation screen, based on systematic evaluation and selection of key parameters, that can be applied to any microbe. Facilitated by a custom-built fully-programmable electroporator, we successfully identified electroporation protocols for eight non-model bacteria using this screen alone. We then combined this electroporation screen with our pooled POSSUM plasmid library to simultaneously evaluate 408 experimental conditions per organism and identified both a protocol and a replicating plasmid for five non-model Proteobacteria spanning three major classes (alpha-, beta-, and gamma-). We report the first electroporation protocols for Shewanella indica, Shewanella putrefaciens 200, Shewanella putrefaciens 95, Halomonas elongata, Piscinibacter sakaiensis, and Duganella zoogloeoides, as well as multiple alternative protocols for Shewanella amazonensis, Shewanella oneidensis, Azospirillum brasilense, Cupriavidus necator, Pseudomonas alcaliphila, and Escherichia coli. Finally, we developed an active learning pipeline to guide the selection of parameters based on gathered experimental data. Using our robotic platform, we iteratively tested 538 conditions over three iterations to improve electroporation for the emerging industrial chassis C. necator, achieving 8.6-fold higher transformation efficiency than state of the art. This work establishes a discovery platform for DNA delivery to diverse and recalcitrant microbes that can be extended broadly to non-model organisms in our biosphere.