ISP Platform powered by Geneformer: Framework for Cross-Species, Sequential, and Multi-Gene In Silico Perturbation Screens with Application to iPS Cell State Transitions
- 1Life Science Center for Survival Dynamics, Tsukuba Advanced Research Alliance (TARA), University of Tsukuba, Tsukuba, Ibaraki, 305-8577, Japan
Abstract
In silico perturbation (ISP) enables virtual genetic screens by simulating targeted genetic manipulations in cells, prioritizing candidate genes before experiments. However, executing ISP requires programming expertise, limiting its use by experimental biologists. Here, we present the ISP Platform (https://github.com/YuyaSanaki/ISP-Platform, version 1.0.2), which automates ISP powered by Geneformer without code execution. The platform enables cross-species analyses in both directions by independently selecting input species and human or mouse Geneformer models. It also supports multi-gene ISP and implements State-feedback ISP, which predicts the gene expression ranking after each perturbation and feeds it into the next one. In single-cell transcriptomes of human and mouse induced pluripotent stem cells (iPSCs), genome-wide knockdown screens prioritized core pluripotency regulators with both species-matched and swapped models. In silico overexpression of seven pluripotency genes shifted somatic cells toward iPSCs more than the Yamanaka factors did. State-feedback ISP simulated all 24 orders of the Yamanaka factors and separated a factor-specific effect in every order, largest when POU5F1 or SOX2 came first. A pairwise in silico overexpression screen nominated candidate modulators of the primed-to-naive transition beyond the NANOG and KLF2 benchmark. The ISP Platform makes Cross-species, State-feedback, and Multi-gene ISP available through a graphical user interface to guide experimental validation.
Related
- scDisentangle: sequential disentanglement for single-cell perturbation response prediction
- RevPert: predicting candidate drivers of transcriptomic state transitions via gallery-native reverse perturbation
- A purinergic receptor dominance model for microglial state transitions: balancing perturbation and restraint during neuroinflammation
- Cause-excluded Effect Evaluation: a framework for the faithful assessment of genetic perturbation target identification
- The Sequential Threshold Model: A Unified Framework for Microbiome-Driven Periodontitis Progression
BibTeX
@misc{sanaki2026platform,
title = {{ISP Platform powered by Geneformer: Framework for Cross-Species, Sequential, and Multi-Gene In Silico Perturbation Screens with Application to iPS Cell State Transitions}},
author = {Y. Sanaki and S. Shigenobu and R. Niwa},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.17.752497},
url = {https://www.biorxiv.org/content/10.64898/2026.09.17.752497v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - ISP Platform powered by Geneformer: Framework for Cross-Species, Sequential, and Multi-Gene In Silico Perturbation Screens with Application to iPS Cell State Transitions
AU - Y. Sanaki
AU - S. Shigenobu
AU - R. Niwa
PY - 2026
DA - 2026/09/17
PB - bioRxiv
DO - 10.64898/2026.09.17.752497
UR - https://www.biorxiv.org/content/10.64898/2026.09.17.752497v1
AB - In silico perturbation (ISP) enables virtual genetic screens by simulating targeted genetic manipulations in cells, prioritizing candidate genes before experiments. However, executing ISP requires programming expertise, limiting its use by experimental biologists. Here, we present the ISP Platform (https://github.com/YuyaSanaki/ISP-Platform, version 1.0.2), which automates ISP powered by Geneformer without code execution. The platform enables cross-species analyses in both directions by independently selecting input species and human or mouse Geneformer models. It also supports multi-gene ISP and implements State-feedback ISP, which predicts the gene expression ranking after each perturbation and feeds it into the next one. In single-cell transcriptomes of human and mouse induced pluripotent stem cells (iPSCs), genome-wide knockdown screens prioritized core pluripotency regulators with both species-matched and swapped models. In silico overexpression of seven pluripotency genes shifted somatic cells toward iPSCs more than the Yamanaka factors did. State-feedback ISP simulated all 24 orders of the Yamanaka factors and separated a factor-specific effect in every order, largest when POU5F1 or SOX2 came first. A pairwise in silico overexpression screen nominated candidate modulators of the primed-to-naive transition beyond the NANOG and KLF2 benchmark. The ISP Platform makes Cross-species, State-feedback, and Multi-gene ISP available through a graphical user interface to guide experimental validation.
ER -