About the data
What the browser contains, how each quantity was derived, and the inferences it does not support.
The dataset
Tahoe-100M: single-cell RNA sequencing of human cancer cell lines, each treated with one small molecule at one of three concentrations, pooled many lines to a plate.
How differential expression is measured
For one drug in one cell line at one concentration:
- The cells are pooled into a single expression profile per well.
- That well is compared against every untreated DMSO control well of the same cell line, with the plate accounted for, so a plate effect is not read as a drug effect.
- A gene counts as changed when it clears a 5% false-discovery rate. "Genes changed" everywhere on this site is that count.
- The direction and size of each gene's change is its log2 fold change: +1 means twice as much, −1 means half as much.
Nothing on this site compares a well against a DMSO control well from a different cell line, and nothing is compared across plates.
How transcription factor activity is inferred
Counting genes says how much happened, not what. For that, each gene's change is mapped onto the transcription factors known to control it, using CollecTRI regulons, giving one score per factor per measurement. A positive score means the factor's target genes went up together, which is what a factor being switched on looks like in RNA.
Factors outside these groups are still scored and still appear in results, under "other".
Genotype annotation
What this cannot tell you
- Not potency. Genes changed is a count of what moved, not an IC50. A drug can kill cells while changing few genes, and change thousands without killing anything.
- Not mechanism, by itself. A pathway score moving tells you that pathway's output changed. It does not establish that the drug acted through it.
- Not a clean causal claim from a genotype split. When mutant lines respond more strongly, the mutation is one of many things those lines differ in, tissue of origin included.
- No growth, viability or protein measurement. RNA only.