betaFeedback

WorksAgricultural and Biological Sciences

S1Pert: Scalable Target-Specific Gene Response Prediction with System-One Intelligence

Abstract

Predicting transcriptional responses to unseen genetic perturbations is essential for extending CRISPR screens across vast target–gene combinations. However, models can achieve strong aggregate predictive performance by reproducing historical gene-level response frequencies without adequately distinguishing the effects of different perturbations. Reliable prediction at the transcriptome scale is further challenged by uncertain response labels and redundant computation across millions of target–gene pairs. We present S1Pert, a System-One framework for target-specific gene response prediction with non-autoregressive online inference. S1Pert combines directional Welch tests and equivalence testing to distinguish statistically supported responses from inconclusive observations. A frozen System-One encoder extracts reusable semantic representations of perturbation targets, query genes, and historical evidence. Compact pair-specific interactions are then computed and cached offline. A lightweight hierarchical cross-attention decoder integrates these representations with cross-target pairwise ranking and synchronized historical-evidence dropout to reduce reliance on gene-level response frequencies. Extensive experiments demonstrate that S1Pert improves target-specific response discrimination while maintaining reliable prediction and efficient inference across diverse cellular contexts. These findings highlight the potential of System-One decision learning for scalable perturbation-response prediction beyond historical gene-level response tendencies. The source code is publicly available at https://github.com/liujiawen-jpg/S1Pert.

Paper details

vlnc:W_e4u8unyb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Agricultural and Biological Sciences

Not recorded for this paper

View statement
Looking for related papers…