RELAY: Relational In-Context Learning for Cellular Perturbation Prediction
- 1Shanghai Jiao Tong University
Abstract
Cellular responses to perturbations vary across multiple experimental variables, such as perturbation identity, cell type, donor, and dose, defining a multidimensional condition space in which each measured population occupies one coordinate. Existing methods predict unmeasured conditions using explicit representations of these variable values, generalizing unseen values therefore depends on the availability and quality of such representations. We present RELAY, a relational in-context model that predicts unmeasured coordinates from measured responses at related coordinates. For each target, RELAY uses measured control--perturbed pairs from conditions that share one or more experimental variables with the target, such as the same perturbation in another cell type. The model receives only which variables are shared, not their values, and uses these demonstrations to predict the target response. We evaluate RELAY on chemical, genetic, and cytokine perturbation datasets under two settings: unseen combinations, where test values are individually observed during training but not in their target combinations, and unseen variable values, where test values are introduced only through context at inference. RELAY outperforms existing in-context methods on unseen variable values and supervised perturbation-prediction methods on unseen combinations. All resources will be open sourced, including code and data.
Related
- The recoverable resolution of cellular perturbation-response prediction
- PertMind: Eliciting Emergent Biological Reasoning in LLM via Reinforcement Learning on Cellular Perturbation Data
- SPECTRA: predicting cellular perturbation responses with Graph Learning over Gene Regulatory Networks
- MetExPred: A Comprehensive Prediction Framework with Protein-Context-Aware Multi-view Learning for Drug Metabolism and Excretion
- MultiFlow: coupled flow matching for predicting single-cell multiomic perturbation responses in unseen cellular contexts
BibTeX
@misc{sun2026relay,
title = {{RELAY: Relational In-Context Learning for Cellular Perturbation Prediction}},
author = {Y. Sun and J. Feng and Y. Wang and W. Xie and Y. Zhang and X. Zhang},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.31.763027},
url = {https://www.biorxiv.org/content/10.64898/2026.10.31.763027v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - RELAY: Relational In-Context Learning for Cellular Perturbation Prediction
AU - Y. Sun
AU - J. Feng
AU - Y. Wang
AU - W. Xie
AU - Y. Zhang
AU - X. Zhang
PY - 2026
DA - 2026/10/31
PB - bioRxiv
DO - 10.64898/2026.10.31.763027
UR - https://www.biorxiv.org/content/10.64898/2026.10.31.763027v1
AB - Cellular responses to perturbations vary across multiple experimental variables, such as perturbation identity, cell type, donor, and dose, defining a multidimensional condition space in which each measured population occupies one coordinate. Existing methods predict unmeasured conditions using explicit representations of these variable values, generalizing unseen values therefore depends on the availability and quality of such representations. We present RELAY, a relational in-context model that predicts unmeasured coordinates from measured responses at related coordinates. For each target, RELAY uses measured control--perturbed pairs from conditions that share one or more experimental variables with the target, such as the same perturbation in another cell type. The model receives only which variables are shared, not their values, and uses these demonstrations to predict the target response. We evaluate RELAY on chemical, genetic, and cytokine perturbation datasets under two settings: unseen combinations, where test values are individually observed during training but not in their target combinations, and unseen variable values, where test values are introduced only through context at inference. RELAY outperforms existing in-context methods on unseen variable values and supervised perturbation-prediction methods on unseen combinations. All resources will be open sourced, including code and data.
ER -