3/11, LSBM Monday Seminar Series #25, Koseki Kobayashi, Bridging Microscopy and Genomics with Machine Learning

Title: Bridging Microscopy and Genomics with Machine Learning

Presenter: Koseki Kobayashi

Affiliation: Postdoctoral Researcher @ Broad Institute of MIT and Harvard + Massachusetts Institute of Technology





Single-cell RNA-seq and other profiling assays have opened new windows into understanding cells' properties, regulation, dynamics, and function at unprecedented resolution and scale. However, these assays are inherently destructive, precluding us from tracking their temporal dynamics. Here, we present Raman2RNA (R2R), an experimental and computational framework to infer single-cell expression profiles in live cells through Raman microscopy images and domain translation using adversarial training. We demonstrate R2R in reprogramming mouse fibroblasts or differentiating mouse embryonic stem cells and show that their expression profiles can be accurately predicted in live cells. We also apply our method to other modalities, providing genomic interpretability to widely available modalities such as H&E stains. Our method should imply broad applications to understanding expression dynamics at scale in vitro and in vivo.