Deep learning-based clustering for endotyping and post-arthroplasty response classification using knee osteoarthritis...

Area of Science:

• Biomarker discovery
• Precision medicine
• Osteoarthritis research

Background:

• Knee osteoarthritis (KOA) is a complex disease with varied patient responses to treatment.
• MicroRNAs and metabolites in biofluids offer potential insights into KOA heterogeneity.
• Multimodal deep learning can integrate complex, high-dimensional omic data.

Purpose of the Study:

• To develop a multimodal deep learning framework for clustering KOA patients based on multiomic data from plasma, synovial fluid, and urine.
• To identify distinct KOA endotypes using integrated microRNA and metabolite profiles.
• To classify patient responses to total knee arthroplasty (TKA) pain and function outcomes within identified clusters.

Main Methods:

• Analyzed microRNA sequencing and metabolomics data from plasma, synovial fluid, and urine of 414 KOA patients.
• Employed a multimodal deep learning variational autoencoder with K-means clustering to integrate 4 high-dimensional datasets.
• Utilized an integrative machine learning framework to classify TKA pain/function responses (WOMAC scores) based on multiomic and clinical data.

Main Results:

• Clustering identified 3 distinct patient groups (endotypes) based on integrated multiomic data.
• Each cluster exhibited unique feature signatures of microRNAs and metabolites, with associated distinct pathway analyses.
• Integrating multiomic data with clinical information significantly improved the classification accuracy of TKA pain/function responses compared to single-domain analyses.

Conclusions:

• A novel multimodal deep learning model effectively integrates multi-biofluid, multiomic data to reveal distinct KOA patient endotypes.
• This approach enhances the classification of TKA surgical outcomes.
• The findings support the potential of precision medicine strategies for personalized KOA treatment and surgical management.