Multikano: An Automatic Cell Type Annotation Tool for Single-Cell Multi-Omics Data Based on Kolmogorov–Arnold Network and Data Augmentation

MultiKano is an advanced tool for automatic cell type annotation in single-cell multi-omics data, integrating transcriptomic and chromatin accessibility data. It uses a novel data augmentation strategy and the Kolmogorov–Arnold Network (KAN) to improve model generalization. Evaluated on six paired single-cell multi-omics datasets, MultiKano outperformed existing methods, particularly in noisy and sparse datasets, demonstrating superior accuracy. Its robustness was confirmed through various experiments, including comparisons with conventional machine learning methods and ablation studies. MultiKano effectively handles biological and technical variability, showing improved accuracy over other methods like TOSICA, and its predictions align well with functional characteristics in downstream analyses. The study suggests further improvements, such as better integration of multi-omics data and leveraging unpaired datasets for enhanced model training.