Learning From Large-Scale (Single-Cell) ‘Omics’
summary
Argues that mechanistic bursty transcription explains the negative binomial counts and that common pipelines like depth normalization plus log or SCTransform often remove real biological variance. Describes a reproducible Cell Atlas framework that reprocesses 526 datasets via an automated GitHub pipeline and an isoform atlas assembled from 38 billion reads. Presents MONOD and the Python package Mono that uses spliced/unspliced matrices to estimate technical versus mechanistic biological variance (works on 10x but is computationally intensive).
AI-written summary · about