scKDGM: KAN-guided Dynamic Graph Masked Learning for Single-Cell RNA-seq Clustering

2026-06-26

AI Summary

Single-cell RNA sequencing data pose significant challenges for robust cell representation and graph construction due to high dimensionality, sparsity, dropout events, and noise, which ultimately hinder clustering performance. To address these issues, this work proposes scKDGM, a novel framework that introduces Kolmogorov–Arnold Networks (KANs) into single-cell clustering for the first time. scKDGM generates multi-view representations through graph-aware gene masking perturbations and employs a KAN-enhanced graph convolutional encoder to learn expressive cell embeddings. It further incorporates a mask-guided expression recovery mechanism that dynamically refines the cell graph topology, moving beyond conventional fixed k-nearest neighbor graphs and static reconstruction paradigms. By integrating cross-view contrastive learning with a zero-inflated negative binomial loss, scKDGM consistently outperforms ten baseline methods across twelve real-world datasets, achieving state-of-the-art results in both Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI).

Abstract

Single-cell RNA sequencing (scRNA-seq) clustering is essential for identifying cell types, but high dimensionality, sparsity, dropout, and technical noise hinder robust expression representation and cell graph construction. Existing masked autoencoders mainly use expression recovery for feature reconstruction, while graph clustering methods usually depend on fixed KNN graphs and do not feed recovered expression back into graph optimization. We propose scKDGM, a KAN-guided dynamic graph masked learning framework for scRNA-seq clustering. scKDGM uses graph-aware distribution preserving gene masking (GDP-Mask) to perturb cell identity, a KAN-based TAKGCN encoder to learn masked-view representations, mask-guided expression recovery to construct a dynamic graph, and cross-view contrastive learning to transfer recovery signals into topology updates. A ZINB loss models overdispersion and zero inflation. Experiments on 12 real scRNA-seq datasets show that scKDGM outperforms 10 baselines in average NMI and ARI.

Problem

• scRNA-seq clustering
• high dimensionality
• sparsity
• dropout
• technical noise

Innovation

• KAN-guided
• dynamic graph
• masked learning
• GDP-Mask
• cross-view contrastive learning