Simultaneous estimation of gene regulatory network structure and RNA kinetics from single cell gene expression

Cells respond to environmental and developmental stimuli by changing their transcriptomes through both regulation of transcription rate and regulated mRNA decay. These biophysical properties determine the transcriptional state of a cell, but measuring them requires metabolic RNA labeling (e.g. 4-thiouracil pulse-chase) to separate RNA decay from synthesis rates. We approach this problem by sequencing individual Saccharomyces cerevisiae cell transcriptomes by continuously sampling from a population without metabolic labeling. Using this continuous-sampling system, we measure expression in 180,000 individual cells both prior to and in response to rapamycin treatment. The rates of change for each transcript can be calculated on a per-cell basis to give smooth, biologically relevant, estimates of RNA velocity. We then train deep learning models to use this transcriptomic and velocity information to make time-dependent predictions about RNA biophysics, and to infer causal regulatory relationships between transcription factors and their genes. Overall design: To minimize the interval between observed data points, we continually pump culture into excess saturated ammonium sulfate to collect cells, fixing transcriptomes at the time of collection. We use this system to quantify the transcriptomic response to rapamycin on a continuous basis, collecting 173,361 cells across 8 defined 10-minute time intervals, from two biological replicate experiments. Samples were collected continuously into 50 mL conicals containing 20 mL RNAlater. Sample collection conicals were changed every 10 minutes and the collected samples were placed on ice. After 20 minutes, 200 ng/mL rapamycin was added to the stirring culture. Sampling continued for 60 additional minutes, yielding 2 10-minute sample pools without rapamycin treatment and 6 10-minute sample pools with rapamycin treatment.