ScNucAdapt: Partial domain adaptation enables cross domain cell type annotation between scRNA-seq and snRNA-seq
Xiran Chen, Quan Zou, Qinyu Cai, Xiaofeng Chen, Weikai Li*, Yansu Wang*
PLoS Computational Biology 2026
Single-cell and single-nucleus RNA sequencing are two powerful technologies that allow scientists to study gene activity in individual cells. However, comparing data between these methods remains challenging because they capture different parts of the cell and are often collected under different conditions. This makes it difficult to consistently identify cell types across experiments, hindering our understanding of health and disease. We developed ScNucAdapt, a computational framework that can automatically transfer cell type knowledge between these two types of datasets, even when they come from different laboratories or tissue conditions. Our method learns to recognize shared patterns while ignoring dataset-specific differences. Through testing on diverse tissues, including bladder, kidney, tumors, and brain, we show that ScNucAdapt consistently outperforms existing approaches. By enabling reliable integration of single-cell and single-nucleus data, our work helps researchers build more complete pictures of cellular diversity across tissues and disease states. This capability is particularly valuable for studying archived frozen samples or fragile cell types that are difficult to analyze with conventional methods, potentially accelerating discoveries in various fields.
ScNucAdapt/Loss: CS Divergence LossScNucAdapt/Network: Main NetworkScNucAdapt/Utils: Utils for ScNucAdaptScNucAdapt/main: main training for ScNucAdapt
Input your preprocessed numpy array data into array(source) and array2(target) in main.py
