Journal: bioRxiv

Article Title: PEPR-GNN: Perturbation-Enhancer-Promoter-RNA Graph Neural Networks for Multiome Perturb-Seq modeling of regulomes

doi: 10.64898/2026.05.05.722311

Figure Lengend Snippet: A. (top) Regulome model illustrating the action of GHMT factors binding enhancers to induce changes in promoter accessibility and gene expression. (bottom) Graphical model represention. B. Mouse embryonic fibroblasts (MEFs) were transduced with a lentiviral library for combinatorial induction of GHMT. Multiome Perturb-Seq readouts for each cell include perturbation, open chromatin, and expression information. C. Genome browser snapshot near the Tnnt2 gene, with regulatory elements labeled. D. Example perturbation-enhancer associations for the Tnnt2 regulome: (left) Open chromatin profiles for cells with detection of all 4 GHMT factors or no factors; (right) 2×2 contingency tables illustrating the calculation of associations. E. A subset of the Tnnt2 regulome after statistical associations, focusing on one perturbation (GHMT) and three enhancers. F. Regulome schematic for systematic testing. G. Genome-wide statistical association analysis for perturbation-enhancer edges. Density of points is indicated. H. Genome-wide statistical association analysis for enhancer-promoter edges. I. Genome-wide statistical association analysis for promoter-RNA edges. J. Example regulome architectures for cardiac gene Tnnt2. K. Example regulome architectures for fibroblast gene Col1a1.

Article Snippet: GHMT TFs were synthesized by Twist Bioscience into a lentiviral vector driven by EF1a, followed by lentivirus production with HEK293T cells as described using packaging vectors psPAX2 and pMD2.G .

Techniques: Binding Assay, Gene Expression, Transduction, Expressing, Labeling, Genome Wide

Journal: bioRxiv

Article Title: PEPR-GNN: Perturbation-Enhancer-Promoter-RNA Graph Neural Networks for Multiome Perturb-Seq modeling of regulomes

doi: 10.64898/2026.05.05.722311

Figure Lengend Snippet: A. Pooled model outputs can be used as features for tSNE embedding, placing every graph on a single manifold organized by features important to the model. B. (top left) Shown is a 2-dimensional tSNE embedding of gene graphs, colored according to GHMT response. The locations of notable genes are indicated. (right and bottom) Regulome models corresponding to key cardiac and fibroblast genes. C. Gene graphs are colored according to the minimum number of GHMT factors needed to up-regulate the gene. The cardiac genes on the right of the embedding can be robustly induced by even a single GHMT factor, and are “easier” to reprogram. D. Gene graphs are colored according to the minimum number of GHMT factors needed to down-regulate the gene. The fibroblast genes on the left of the embedding require the induction of many GHMT factors to repress gene expression, and are “harder” to reprogram. E. An example where Tbx5 exhibits contextual regulation of Ttn, Tmem71, and Hcn4 depending on the context of GHM factors expressed in a cell. In contrast, Tbx5 does not exhibit contextual regulation of Actb and Des. F. Gene graphs are colored according to the number of TFs that exhibit contextual up-regulation. The graphs on the top of the embedding exhibit the most contextual regulation.

Article Snippet: GHMT TFs were synthesized by Twist Bioscience into a lentiviral vector driven by EF1a, followed by lentivirus production with HEK293T cells as described using packaging vectors psPAX2 and pMD2.G .

Techniques: Gene Expression