scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing
Researchers have developed scLLM-DSC, a new framework that enhances deep structural clustering for single-cell RNA sequencing data by integrating Large Language Model (LLM) knowledge. This method addresses the limitations of existing numerical pattern-focused approaches by incorporating biological function and semantic information from genes. scLLM-DSC creates a unified latent space by combining a knowledge-driven semantic view with a structure-aware topological view, using a cross-modal contrastive alignment mechanism. Benchmarks show scLLM-DSC surpasses eleven state-of-the-art methods in clustering accuracy. AI
IMPACT This research introduces a novel method for analyzing biological data, potentially accelerating discoveries in genomics and cell biology.