VirtualSkin integrates skin-cell identity with intervention-response prediction
- 1Center for Computational Dermatology, Northbridge Institute of Biomedical AI, Boston, Massachusetts, USA; Department of Biomedical Informatics, Halcyon Universi
Abstract
Human skin atlases resolve a diverse set of cellular states, yet connecting these states to their responses to intervention remains difficult across independently sampled tissues. Here we present VirtualSkin, a cell language model that aligns skin transcriptomes with biological descriptions and supports intervention conditioned response prediction. We evaluated the model across independent atlas, disease, anatomical and perturbation cohorts, with donors, studies and perturbations withheld according to the experimental design. VirtualSkin improved inflammatory skin annotation by 0.056 macro F1 points and cell text retrieval by 0.12 Recall@10 points relative to the strongest corresponding comparators. The representation retained cell identity during study integration and transferred across ageing, systemic sclerosis, lymphoma and anatomical sites. Across five intervention cohorts, response prediction improved by 0.03 0.09 response-score points, with agreement at the levels of gene expression change, gene recovery and transcriptional programs. Paired tissue trajectories provided a complementary assessment in two donors per intervention. Component ablations and pairing controls localized the predictive contributions, while uncertainty estimates supported selective prediction under distribution shift. These results connect skin-cell recognition to intervention response prediction within a common evaluation framework and define the conditions under which predictions can inform experimental prioritization.
Related
- From pose to behavior: SABER integrates identity-resolved multi-animal pose tracking with language-model-based behavioral factor discovery
- A deep-learning microglial response score integrates single-cell and spatial transcriptomics for Alzheimer’s disease stratification
- CLSP-REQA: a real-time quality-aware closed-loop seizure prediction framework with Mamba-BiLSTM and confidence-gated intervention
- Determinants of gut microbiome composition and its response to a dietary intervention in a multi-ethnic cohort: a pop-up village study
- Phage intervention improves colitis and response to corticosteroids by attenuating virulence of Crohn’s disease–associated bacteria
BibTeX
@misc{velland2026virtualskin,
title = {{VirtualSkin integrates skin-cell identity with intervention-response prediction}},
author = {M. Velland and E. Norwick and S. Ardene and T. Marwick and L. Corven and A. Solmere and N. Calder and R. Quillon and I. Renn and M. Elian},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.06.756877},
url = {https://www.biorxiv.org/content/10.64898/2026.10.06.756877v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - VirtualSkin integrates skin-cell identity with intervention-response prediction
AU - M. Velland
AU - E. Norwick
AU - S. Ardene
AU - T. Marwick
AU - L. Corven
AU - A. Solmere
AU - N. Calder
AU - R. Quillon
AU - I. Renn
AU - M. Elian
PY - 2026
DA - 2026/10/06
PB - bioRxiv
DO - 10.64898/2026.10.06.756877
UR - https://www.biorxiv.org/content/10.64898/2026.10.06.756877v1
AB - Human skin atlases resolve a diverse set of cellular states, yet connecting these states to their responses to intervention remains difficult across independently sampled tissues. Here we present VirtualSkin, a cell language model that aligns skin transcriptomes with biological descriptions and supports intervention conditioned response prediction. We evaluated the model across independent atlas, disease, anatomical and perturbation cohorts, with donors, studies and perturbations withheld according to the experimental design. VirtualSkin improved inflammatory skin annotation by 0.056 macro F1 points and cell text retrieval by 0.12 Recall@10 points relative to the strongest corresponding comparators. The representation retained cell identity during study integration and transferred across ageing, systemic sclerosis, lymphoma and anatomical sites. Across five intervention cohorts, response prediction improved by 0.03 0.09 response-score points, with agreement at the levels of gene expression change, gene recovery and transcriptional programs. Paired tissue trajectories provided a complementary assessment in two donors per intervention. Component ablations and pairing controls localized the predictive contributions, while uncertainty estimates supported selective prediction under distribution shift. These results connect skin-cell recognition to intervention response prediction within a common evaluation framework and define the conditions under which predictions can inform experimental prioritization.
ER -