Multi-Omics clustering by integrating clinical features from large language model.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics clustering integrates diverse omics data for biological system analysis and cancer subtyping.
• Current methods often neglect the valuable clinical context present in unstructured medical text.

Purpose of the Study:

• To develop a novel framework for multi-omics clustering that integrates clinical features extracted by large language models (LLMs).
• To enhance cancer subtyping by enriching omics data with structured clinical information.

Main Methods:

• Utilized a BERT-based model to extract clinical features from pathology reports, converting unstructured text into structured data.
• Integrated extracted clinical features with omics data using an autoencoder to enrich omics layers.
• Employed singular value decomposition (SVD) for latent subspace projection, followed by spectral clustering for final results.

Main Results:

• The proposed framework demonstrated superior performance in multi-omics clustering for cancer subtyping across six datasets and three omics levels.
• Integration of LLM-derived clinical features significantly improved clustering performance compared to existing methods.
• Validated the framework's effectiveness against several state-of-the-art approaches.

Conclusions:

• Incorporating clinical context derived from LLMs is crucial for advancing multi-omics analysis.
• The proposed framework offers a powerful tool for precision medicine by improving cancer subtyping through enhanced multi-omics clustering.
• Highlights the transformative potential of LLMs in biomedical research and clinical applications.