We built GenoPHI: a machine learning workflow that predicts phage-host interactions at strain level. This could help rapidly select phages to treat drug-resistant bacterial infections or for microbiome engineering without exhaustive lab testing.
www.biorxiv.org/content/10.1...
www.biorxiv.org
OUR APPROACH: Interpretable genomic features + ML to predict interactions. Phylogeny-agnostic feature construction so it works for novel phages and bacteria. We trained and tested across 5 public datasets (128,357 interactions total) and validated with high-throughput phenotyping + RB-TnSeq.
THE CHALLENGE: Bacteria and phages are incredibly diverse. Experimentally testing each phage against a new bacterial target isn't feasible. But finding the right phage quickly could be life-saving, especially for drug-resistant infections where treatment options are limited.
Avery Noonan
Interested? Read the pre-print, or check out our code ⬇️
Preprint: www.biorxiv.org/content/10.1...
GitHub repo: github.com/Noonanav/Gen...
Huge thanks to co-authors at Berkeley Lab, UC Berkeley (Phage Foundry, NSF EDGE): Lucas Morinière, Krish Patel, Melina Pena, Madeline Svab, Alexey Kazakov, Adam Deutschbauer, @vivekmutalik.bsky.social , Adam Arkin & collaborators at Penn State: Edwin Omar Rivera-López, Edward Dudley
