Re-analysis of the Human Transcription Factor Atlas Recovers TF-Specific Signatures from Pooled Single-Cell Screens with Missing Controls

Arka Jain; Umesh Sharma

cs.LGApr 2, 2026

Re-analysis of the Human Transcription Factor Atlas Recovers TF-Specific Signatures from Pooled Single-Cell Screens with Missing Controls

Authors: Arka Jain, Umesh Sharma

Organizations: Mayo Clinic

Abstract

Public pooled single-cell perturbation atlases are valuable resources for studying transcription factor (TF) function, but downstream re-analysis can be limited by incomplete deposited metadata and missing internal controls. Here we re-analyze the human TF Atlas dataset (GSE216481), a MORF-based pooled overexpression screen spanning 3,550 TF open reading frames and 254,519 cells, with a reproducible pipeline for quality control, MORF barcode demultiplexing, per-TF differential expression, and functional enrichment. From 77,018 cells in the pooled screen, we assign 60,997 (79.2%) to 87 TF identities. Because the deposited barcode mapping lacks the GFP and mCherry negative controls present in the original library, we use embryoid body (EB) cells as an external baseline and remove shared batch/transduction artifacts by background subtraction. This strategy recovers TF-specific signatures for 59 of 61 testable TFs, compared with 27 detected by one-vs-rest alone, showing that robust TF-level signal can be rescued despite missing intra-pool controls. HOPX, MAZ, PAX6, FOS, and FEZF2 emerge as the strongest transcriptional remodelers, while per-TF enrichment links FEZF2 to regulation of differentiation, EGR1 to Hippo and cardiac programs, FOS to focal adhesion, and NFIC to collagen biosynthesis. Condition-level analyses reveal convergent Wnt, neurogenic, EMT, and Hippo signatures, and Harmony indicates minimal confounding batch effects across pooled replicates. Our per-TF effect sizes significantly agree with Joung et al.'s published rankings (Spearman $ρ= -0.316$ , $p = 0.013$ ; negative because lower rank indicates stronger effect). Together, these results show that the deposited TF Atlas data can support validated TF-specific transcriptional and pathway analyses when paired with principled external controls, artifact removal, and reproducible computation.

Figures & tables

Panel of Figure 1 — Figure 1: UMAP of the combined TF Atlas dataset. (A) 254,519 cells colored by sample identity, showing separation between experimental conditions and overlap among pooled screen replicates. (B) Leiden clustering identifies 83 transcriptionally distinct populations.

Panel of Figure 2 — Figure 2: MORF barcode demultiplexing of pooled screen samples. (A) UMAP of 77,018 cells colored by demultiplexing status: assigned (blue), ambiguous (orange), undetected (green). (B) Distribution of assigned cells across TFs.

Figure 3: TF-specific differentially expressed genes per TF (vs EB control with background subtraction, $|$ log 2 FC $|$ $>$ 0.5, FDR $<$ 0.05). Red: upregulated; blue: downregulated. Top 30 TFs shown.

TF	Cells	Specific DEGs	Up	Down	One-vs-rest DEGs	Top upregulated genes
HOPX	6,645	2,861	1,882	979	0	HOPX, PAX6, SOX1, VIM
MAZ	4,672	2,585	1,691	894	0	MAZ, TPT1, ENO1
PAX6	3,913	2,412	1,569	843	0	PAX6, SOX1, FGF8
FOS	990	1,128	702	426	136	ANXA1, CDH13, UTF1, EGR1
FEZF2	703	1,033	680	353	173	FOXA2, TFF3, FOXB1, SOX21
NFIC	1,294	914	553	361	117	NUPR1, S100A14, ACTG2

Table 1: Top TFs by number of TF-specific differentially expressed genes from the pooled screen (vs EB control with background subtraction). One-vs-rest DEG counts shown for comparison.

Panel of Figure 4 — Figure 4: Functional enrichment of condition-level DEGs. (A) Most frequently enriched GO/KEGG terms across perturbations, ranked by minimum adjusted $p$ -value. (B) Dotplot showing term enrichment across perturbations (dot size = gene count, color = $-$ log 10 adjusted $p$ -value).

Panel of Figure 5 — Figure 5: Functional enrichment of per-TF DEGs from pooled screen. (A) Most frequently enriched GO/KEGG terms across 13 TFs with significant ORA results, ranked by minimum adjusted $p$ -value. (B) Dotplot showing per-TF enrichment landscape (dot size = gene count, color = $-$ log 10 adjusted $p$ -value). Only TFs with $\geq$ 5 significant DEGs ( $|$ log 2 FC $|$ $>$ 0.5, FDR $<$ 0.05) were tested.

Figure 6: TF-vs-pathway enrichment heatmap. Clustered heatmap of the top 20 most recurrent enriched terms across 11 TFs with significant ORA results. Color intensity represents $-$ log 10 (adjusted $p$ -value), clipped at 10. Hierarchical clustering reveals modules of co-enriched pathways and TFs with similar functional programs.

Figure 7: Batch effects assessment via Harmony integration. UMAP visualizations of the 8 pooled screen replicates before and after Harmony batch correction. Minimal changes confirm that inter-replicate batch effects are negligible and do not confound downstream analyses. Harmony converged in 3 iterations.

Figure 8: Validation of per-TF signatures against Joung et al. (2023) rankings. Scatter plots comparing our per-TF DEG count with Joung et al. average rank (left) and scRNA-seq rank (right). Lower rank indicates stronger TF effect, so the negative Spearman $\rho$ reflects positive agreement between the two studies. The significant correlation ( $\rho=-0.316$ , $p=0.013$ for scRNA-seq rank) validates that our independently derived signatures recapitulate the original study’s findings.

Single-Cell Transcriptomics

Single-cell transcriptomics measures gene activity cell by cell to map cell types, states, interactions, and responses across tissues and conditions.

See all 65 papers

GeneSpeak-FP: Target and Compound Retrieval from Observed Cell-Level Perturbation Signatures
Jul 20, 2026Kseniia Vaniushkina, Jeongmin Lim, Jinyong ParkSingle-Cell Perturbation Modeling Information Retrieval
AIFFEL Research, Modulabs, Republic of Korea
PopPert: Population-level Joint-Distribution Modeling for Single-Cell Perturbation Prediction
Sep 1, 2026Handong Wang, Jiaxin Qi, Haochen Feng +1Single-Cell Perturbation Modeling Single-Cell RNA Sequencing
Computer Network Information Center, Chinese Academy of Sciences · University of Chinese Academy of Sciences
PerturbMap: Cross-Context Transfer of Single-Cell Perturbation Responses
Jul 30, 2026Panpan Cui, Yiqi Liu, Wenhao SunSingle-Cell Perturbation Modeling Cross-Domain Transfer Learning
School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences · Institute of Computing Technology, CAS; University of Chinese Academy of Sciences · 3Hong Kong University of Science and Technology

Jul 20, 2026J/K move · Enter open · S save

GeneSpeak-FP: Target and Compound Retrieval from Observed Cell-Level Perturbation Signatures

Kseniia Vaniushkina, Jeongmin Lim, Jinyong Park

AIFFEL Research, Modulabs, Republic of Korea

Single-Cell Perturbation Modeling Information Retrieval Genomics Molecular Representation Learning Drug Discovery

Open paper AskPDF arXiv

Single-cell perturbation profiles may help identify known compounds across cellular contexts. We present GeneSpeak-FP, a gene-token Transformer that jointly ranks recorded compounds and their target annotations from a treated cell relative to a cell-line-specific DMSO reference, together with a coarse organ token. On Tahoe-100M, we split 11,673 drug--cell-line groups 90/10 within each compound. No complete group is shared across partitions, although compound and cell-line identities may recur. In a 38,400-query Monte Carlo pass, the correct compound appears first for 12.9% of cells and within the top 10 for 34.3%, among 379 fixed candidates. The model retrieves 40.8% of annotated targets within the top 10 and 54.4% within the top 20, from a vocabulary of 278 genes. Top-10 compound retrieval and target recall are approximately 13 and 11 times their expected values under uniformly random rankings; top-1 compound retrieval is nearly 49 times chance. Top-10 compound retrieval also exceeds the strongest evaluated post-hoc bag-of-genes control by more than nine times. A separate stochastic pass from the same validation pool gives nearly identical aggregate results. These findings support retrospective candidate-space narrowing in held-out drug--cell-line contexts. Because compound identities may recur and target labels are non-exhaustive metadata, they do not establish unseen-compound discovery, causal target identification, or clinical utility.

Abstract

Single-Cell Transcriptomics

Explore similar work

GeneSpeak-FP: Target and Compound Retrieval from Observed Cell-Level Perturbation Signatures

PopPert: Population-level Joint-Distribution Modeling for Single-Cell Perturbation Prediction

PerturbMap: Cross-Context Transfer of Single-Cell Perturbation Responses