A common genetic architecture enables the lossy compression of large CRISPR libraries
Zhao, B.; Rao, Y.; Gilbert, L.; Pritchard, J. R.
Show abstract
There are thousands of ubiquitously expressed mammalian genes, yet a genetic knockout can be lethal to one cell, and harmless to another. This context specificity confounds our understanding of genetics and cell biology. 2 large collections of pooled CRISPR screens offer an exciting opportunity to explore cell specificity. One explanation, synthetic lethality, occurs when a single "private" mutation creates a unique genetic dependency. However, by fitting thousands of machine learning models across millions of omic and CRISPR features, we discovered a "public" genetic architecture that is common across cell lines and explains more context specificity than synthetic lethality. This common architecture is built on CRISPR loss-of-function phenotypes that are surprisingly predictive of other loss-of-function phenotypes. Using these insights and inspired by the in silico lossy compression of images, we use machine learning to identify small "lossy compression" sets of in vitro CRISPR constructs where reduced measurements produce genome-scale loss-of-function predictions.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Partial gene suppression improves identification of cancer vulnerabilities when CRISPR-Cas9 knockout is pan-lethal 98%
- Conditional resampling improves calibration and sensitivity in single cell CRISPR screen analysis 97%
- Enhancer regulatory networks globally connect non-coding breast cancer loci to cancer genes 97%
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.