Stable-Shift: Biologically Structured Prediction of Transcriptional Responses to Unseen Gene Perturbations Conference: The 17th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics; June 30–July 3, 2026; Rende, Italy CCS: Applied computing Bioinformatics CCS: Applied computing Computational genomics CCS: Computing methodologies Machine learning CCS: Computing methodologies Neural networks
Sajib Acharjee Dip, Liqing Zhang · 2026
Read Stable-Shift: Biologically Structured Prediction of Transcriptional Responses to Unseen Gene Perturbations Conference: The 17th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics; June 30–July 3, 2026; Rende, Italy CCS: Applied computing Bioinformatics CCS: Applied computing Computational genomics CCS: Computing methodologies Machine learning CCS: Computing methodologies Neural networks by Sajib Acharjee Dip, Liqing Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-24940.txt.
Description
Sajib Acharjee Dip and Liqing Zhang present Stable-Shift, a method that predicts how gene expression responds to perturbations of genes never tried in training. Extrapolation to unseen genes is the hard part. Stable-Shift first pools single cell measurements into perturbation level expression shifts. It then fits a low rank basis of responses using only training perturbations. An unseen gene's coordinates in that basis are predicted from biological context, such as interaction networks and Gene Ontology annotations.
Graph convolution integrates the context inputs in the evaluated version. On the supplied K562 Perturb-seq benchmark, Stable-Shift reaches cosine similarity of about 0.59 against measured responses. The GEARS baseline scores about 0.57. Spearman correlation and top gene precision also favour Stable-Shift among the methods compared. Across five unseen gene splits its mean score stays near 0.589 with small spread.
Catalog record
| CREATOR | Sajib Acharjee Dip, Liqing Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.24940