scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

scLLM-DSC is a novel LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering framework designed for single-cell RNA sequencing (scRNA-seq) analysis. It addresses the semantic agnosticism of traditional clustering methods and the structural mismatch when directly adapting Large Language Models (LLMs) for discriminative tasks. The framework establishes a semantically-grounded representation by combining a Knowledge-Driven Semantic View, derived from NCBI gene priors and contextualized Cell2Sentence embeddings, with a Structure-Aware Topological View extracted via a graph-guided encoder. A crucial cross-modal contrastive alignment mechanism enforces consistency between biological semantics and transcriptomic features in a unified latent space. Extensive benchmarks demonstrate scLLM-DSC significantly outperforms eleven state-of-the-art baselines, achieving 88.80% ACC, 85.35% NMI, and 83.04% ARI, while also providing explicit biological attribution.

Key takeaway

For Research Scientists or Machine Learning Engineers analyzing single-cell RNA sequencing data, if you are struggling with the semantic interpretability or accuracy of current clustering methods, you should consider adopting a knowledge-enhanced approach. scLLM-DSC demonstrates that integrating LLM-derived biological semantics with structural transcriptomic features via cross-modal alignment significantly improves clustering fidelity and provides explicit biological attribution. You can achieve more robust and interpretable cell population identification by moving beyond purely numerical patterns.

Key insights

Integrating LLM-derived biological semantics with structural transcriptomic features enhances scRNA-seq clustering accuracy and interpretability.

Principles

• Semantic grounding improves biological reasoning.
• Cross-modal alignment unifies diverse data views.
• Task-specific optimization overcomes general model limitations.

Method

Collect NCBI gene metadata, encode with LLM, create dual-path cell semantic embeddings, extract structural features via graph encoder, then align and fuse these views using contrastive learning for clustering.

In practice

• Use LLMs as semantic mappers for biological priors.
• Combine expression-weighted and sequence-based gene semantics.
• Apply InfoNCE loss for cross-modal feature alignment.

Topics

• Single-Cell RNA Sequencing
• Deep Structural Clustering
• Large Language Models
• Cross-Modal Alignment
• Biological Semantics
• Cell Population Identification

Code references

• XPgogogo/scLLM-DSC

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing — scLLM-DSC integrates LLM-derived biological semantics with transcriptomic features for superior scRNA-seq clustering.
• Benchmarking large language models for cell-free RNA diagnostic biomarker discovery — Large language models show promise for diagnostic biomarker discovery, particularly in gene panel nomination.
• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.
• Language Models as Semantic Teachers: Post-Training Alignment for Medical Audio Understanding — Aligning audio encoders with medical LLMs via a "semantic teacher" paradigm significantly enhances clinical diagnostic performance.
• Harnessing the power of single-cell large language models with parameter-efficient fine-tuning using scPEFT — scPEFT efficiently adapts single-cell LLMs to new tasks, reducing computational demands and improving performance.
• REVEAL: Multimodal Vision-Language Alignment of Retinal Morphometry and Clinical Risks for Incident AD and Dementia Prediction — Jointly modeling retinal images and clinical risk factors via VLM improves early AD and dementia prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.