Clinical-transcriptomic classification of lumbar disc degeneration enhanced by machine learning

Area of Science:

• Biochemistry
• Genomics
• Orthopedics

Background:

• Lumbar disc degeneration (LDD) exhibits significant heterogeneity in clinical and pathological aspects.
• The relationship between LDD's transcriptomic variations and its clinical heterogeneity remains unclear.
• This study aimed to classify LDD transcriptomically and assess if clinical features predict molecular subtypes.

Purpose of the Study:

• To classify degenerated discs in LDD patients based on transcriptomic profiles.
• To determine if molecular subtypes of LDD can be predicted using clinical features.
• To understand the molecular mechanisms underlying LDD subtypes.

Main Methods:

• Bulk RNA sequencing of 122 nucleus pulposus tissues from 108 LDD patients.
• Unsupervised clustering to analyze RNA-seq data and identify transcriptional signatures.
• Integration of bulk and single-cell sequencing for cell subpopulation analysis.
• Machine learning models developed to correlate molecular classification with clinical features.

Main Results:

• LDD was classified into four subtypes (C1-C4) with distinct molecular signatures and extracellular matrix (ECM) remodeling.
• C1 showed collagenesis (via TRPV4, PIEZO1), C2 exhibited chondrogenic/osteogenic phenotypes, C3 had reduced chondrogenesis (disrupted hypoxia), and C4 involved macrophages and fibrogenesis (TNF-α).
• A random forest model accurately stratified LDD molecular subtypes using 12 clinical features (ROC: 0.9312, accuracy: 0.84).

Conclusions:

• Four distinct molecular subtypes of LDD were identified.
• These molecular subtypes can be accurately stratified using clinical features.
• This classification facilitates precise diagnostics and personalized treatment strategies for LDD.