Generalized inference of noise-corrected state variability in single cells
Single-cell omics methods operate at their detection limit and are less conducive to technical controls than mini-bulk approaches, which combine cells of the same type before profiling. To integrate mini-bulk samples with paired split-pool controls more holistically, we built stochprofIC, an algorithm that considers data, noise, and covariates in a distributional-regression framework. We vet…
The article discusses the development of stochprofIC, an algorithm designed to enhance single-cell omics methods. These methods are known for their detection limits and technical challenges compared to mini-bulk approaches, which involve combining cells of the same type before profiling. To address this, stochprofIC integrates mini-bulk samples with paired split-pool controls using a distributional-regression framework that accounts for data, noise, and covariates.
The authors validated stochprofIC through synthetic datasets and compared it to an adjusted-variance alternative using mini-bulk data from mouse brains and human breast cancers. The noise-batch corrections provided by stochprofIC allow for the inference of the single-cell populations that make up the mini-bulk samples, which have been consistently validated through RNA FISH in various contexts.
These outputs are then used as inputs for a systems model of nucleocytoplasmic transport, enabling the distinction of heterogeneous trafficking states.
One significant application of stochprofIC is the pharmacologic inhibition of a MAP3K11-high state, which was found to inhibit apoptosis in primary breast-cancer organoids. This finding underscores the relevance of the algorithm to phenotypes. Additionally, stochprofIC supports mini-bulking by deconvolving changes in both location and scale, which is crucial when seeking high-sensitivity and reliable profiles.
Rewritten by urgent.news; may contain errors — check the original.
Source: biorxiv.org