L0 · Target-feature discovery T

D'Alessio et al.

D'Alessio AC, Fan ZP, Wert KJ, Baranov P, Cohen MA, Saini JS, Cohick E, Charniga C, Dadon D, Hannett NM, Young MJ, Temple S, Jaenisch R, Lee TI, Young RA

2015 · Stem Cell Reports

Classical Level 0 exemplar. Identifies TFs specifically enriched in a target cell type against a broad background of other lineages.

Abstract

From the original paper, Stem Cell Reports · PubMed

Hundreds of transcription factors (TFs) are expressed in each cell type, but cell identity can be induced through the activity of just a small number of core TFs. Systematic identification of these core TFs for a wide variety of cell types is currently lacking and would establish a foundation for understanding the transcriptional control of cell identity in development, disease, and cell-based therapy. Here, we describe a computational approach that generates an atlas of candidate core TFs for a broad spectrum of human cells. The potential impact of the atlas was demonstrated via cellular reprogramming efforts where candidate core TFs proved capable of converting human fibroblasts to retinal pigment epithelial-like cells. These results suggest that candidate core TFs from the atlas will prove a useful starting point for studying transcriptional control of cell identity and reprogramming in many human cell types.

Summary

The systematic reprogramming-TF identification of D'Alessio and colleagues is the classical Level 0 exemplar. Its defining strategy is to identify transcription factors that are specifically enriched in a target cell type relative to a broad background of other lineages.

This is precisely the Level 0 logic: the method asks which factors define $P_T$ , not which intervention will move $P_S$ toward $P_T$ . The output is a ranked list of candidate identity factors. The intervention $u$ never enters the scoring objective, and no post-intervention state is predicted.

Why this level

Level 0 because the method scores features $A$ against a target/background contrast $(P_T, P_B)$ and returns a Top-K list of candidate identity determinants; the intervention $u$ does not enter the objective and no forward map is involved. Representation family is signature / state-matching because the underlying object is a transcription-factor activity signature derived from a multi-tissue reference. Cited as the classical Level 0 exemplar in §3.1.