scArchon: a scalable benchmarking framework for assessing single-cell perturbation models

Pre-print: https://www.biorxiv.org/content/10.1101/2025.06.23.661046v1

GitHub: https://github.com/hdsu-bioquant/scArchon

Here, you will find the datasets used in the scArchon work. 

scArchon is a modular, reproducible benchmarking platform for evaluating single-cell perturbation response prediction tools. Built on Snakemake, it provides an extensible framework to compare deep learning methods across diverse datasets using both statistical and biological metrics. Why scArchon? While many tools exist to predict single-cell responses to perturbations (e.g., drug treatments), their systematic comparison has been limited. Importantly, scArchon provides environments for each of the tools to aleviate problems related to their installation. scArchon helps standardize benchmarking and highlights important nuances—such as when models with high quantitative scores fail to retain key biological signals. Tools compared: scgen, trvae, scpregran, cellot, cpa, scvidr, screen, scpram, scdisinfact, c2s.