Topology-adjusted knockout (TAKO): interpretable in silico single-gene knockout ranking on wild-type single-cell gene regulatory networks

Abstract

Predicting single-gene knockout effects from wild-type single-cell RNA-seq data remains challenging when perturbation experiments are costly and existing approaches require knockout-specific model training. Here we introduce Topology-Adjusted Knockout (TAKO), a computationally efficient WT-only framework for single-gene KO impact ranking on single-cell gene regulatory networks. The method applies a strict cut-in-and-out intervention on network transition dynamics and quantifies impact by contrasting stationary propagation profiles before and after perturbation under the same restart distribution. In SERGIO simulations, TAKO recovered direct regulator–target structure but did not show a consistent global advantage over a matched absolute Pearson-correlation ranking across network sizes. Direct evaluation against measured responses in Trem2 and Nkx2-1 KO data likewise yielded mixed global discrimination, although TAKO contained more differentially expressed genes among the top-ranked candidates in both examples. These results position TAKO as a fast, interpretable topology-based prioritisation method rather than a generally superior predictor of post-KO expression responses.