When label-free morphology is sufficient for targeted cellular measurements

Label-free morphology can be sufficient for targeted cellular measurements, but its effectiveness relies on the specific application. Researchers evaluated the sufficiency of such measurements using over 10 million paired phase-reporter observations from 1,000 gene knockouts, 52 fluorescent reporters, and 73 A549 screens. Ten predictors exhibited varying recovery rates across different fields, genes, and screens.

The same-cell pairing method revealed that reporter-state information extends beyond average perturbation levels and allows for covariate measurement. While perturbation ranking often remains consistent even when response magnitude variation decreases, predictions for knockout responses still maintain a median 79% of the strongest hits. However, functional-term retention varies.

Using atlas-derived criteria, scientists identified 30 quantitative and 7 ranking proxies for a single ensemble, with reference reproducibility playing a crucial role in interpretation. In an independent study involving primary hepatocytes, brightfield predictions successfully prioritized 10 out of the 11 compounds with the most significant measured metabolic-activity losses among 217 held-out compounds.

Consequently, measurement sufficiency depends on factors beyond predictive correlation alone, including the target, predictor, context, and scientific task.