SubcellPert: Benchmarking Subcellular Perturbation Prediction
- 1The Chinese University of Hong Kong
Abstract
Perturbations change both RNA abundance and intracellular organization, but gene-count predictors omit the positions needed to model spatial responses. To address this limitation, we introduce \textbf{\textit{SubcellPert}}, a benchmark for perturbation-response prediction at single-RNA-molecule resolution. It comprises 14 imaging datasets with 13.88 million cells and 3.65 billion RNA molecules. Given source observations and a directed condition change, models generate target RNA populations in normalized two-dimensional coordinates. Two independently trained tasks separate abundance from localization: \textit{Oracle} predicts coordinates given exact target counts; \textit{Joint} predicts both. Eight baselines span source resampling, classical transport, and frozen foundation-model adapters. Fixed perturbation and cellular-context holdouts evaluate abundance error, target-gene coverage, and conditional spatial error. Through within-dataset analyses, we find context-dependent gains from spatial features in condition decoding. Interpreting spatial responses requires distinguishing changes within spatial clusters from population composition, and RNA fractions from enrichment relative to compartment geometry. Through comprehensive experiments, we find that Gaussian transport achieves the lowest Oracle spatial error on eight of eleven cellular-context OOD datasets. Audits show that better centroid accuracy can accompany worse radial distributions, and that 19 of 190 Joint adapter fits retain source counts after validation. SubcellPert opens a spatial frontier for virtual cell research, laying the groundwork for predictive models that capture both the molecular composition and intracellular organization of cellular responses. The code is available at \url{https://github.com/king159/SubcellPert}.
Related
- AmbiModBench: Benchmarking Gene Perturbation Prediction Beyond Shared Responses
- CellChain: Benchmarking Cross-Scale Evidence Use for Perturbation Reasoning
- Auditing Confidence-Calibrated Benchmarking of Single-Cell Perturbation Models
- PerturbBridge: Conditional Latent Schr\"{o}dinger Bridge for Single-Cell Perturbation Prediction
- Benchmarking Large Language Models for Assay-Aware Drug-Target Binding Affinity Prediction
BibTeX
@misc{wang2026subcellpert,
title = {{SubcellPert: Benchmarking Subcellular Perturbation Prediction}},
author = {J. Wang and H. Cao and X. Dang and T. Zhou and Y. Zhao and C.-Y. Hsieh and C. Gu and P. A. Heng},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.11.758305},
url = {https://www.biorxiv.org/content/10.64898/2026.10.11.758305v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - SubcellPert: Benchmarking Subcellular Perturbation Prediction
AU - J. Wang
AU - H. Cao
AU - X. Dang
AU - T. Zhou
AU - Y. Zhao
AU - C.-Y. Hsieh
AU - C. Gu
AU - P. A. Heng
PY - 2026
DA - 2026/10/11
PB - bioRxiv
DO - 10.64898/2026.10.11.758305
UR - https://www.biorxiv.org/content/10.64898/2026.10.11.758305v1
AB - Perturbations change both RNA abundance and intracellular organization, but gene-count predictors omit the positions needed to model spatial responses. To address this limitation, we introduce \textbf{\textit{SubcellPert}}, a benchmark for perturbation-response prediction at single-RNA-molecule resolution. It comprises 14 imaging datasets with 13.88 million cells and 3.65 billion RNA molecules. Given source observations and a directed condition change, models generate target RNA populations in normalized two-dimensional coordinates. Two independently trained tasks separate abundance from localization: \textit{Oracle} predicts coordinates given exact target counts; \textit{Joint} predicts both. Eight baselines span source resampling, classical transport, and frozen foundation-model adapters. Fixed perturbation and cellular-context holdouts evaluate abundance error, target-gene coverage, and conditional spatial error. Through within-dataset analyses, we find context-dependent gains from spatial features in condition decoding. Interpreting spatial responses requires distinguishing changes within spatial clusters from population composition, and RNA fractions from enrichment relative to compartment geometry. Through comprehensive experiments, we find that Gaussian transport achieves the lowest Oracle spatial error on eight of eleven cellular-context OOD datasets. Audits show that better centroid accuracy can accompany worse radial distributions, and that 19 of 190 Joint adapter fits retain source counts after validation. SubcellPert opens a spatial frontier for virtual cell research, laying the groundwork for predictive models that capture both the molecular composition and intracellular organization of cellular responses. The code is available at \url{https://github.com/king159/SubcellPert}.
ER -