Interpretable multimodal machine learning (IMML) framework reveals pathological signatures of distal sensorimotor...

Area of Science:

• Neurology
• Bioinformatics
• Computational Biology

Background:

• Distal sensorimotor polyneuropathy (DSPN) is a prevalent neurological disorder linked to aging, obesity, and diabetes, contributing to significant morbidity and mortality.
• The multifactorial nature of DSPN and its underlying mechanisms remain incompletely understood.

Purpose of the Study:

• To develop an Interpretable Multimodal Machine Learning (IMML) framework for predicting DSPN prevalence and incidence.
• To leverage IMML's interpretability for identifying novel biomarkers associated with DSPN.
• To analyze deep multimodal data from a large population cohort for DSPN insights.

Main Methods:

• Utilized the KORA F4/FF4 cohort (1091 participants) with comprehensive multimodal data (clinical, genomics, methylomics, transcriptomics, proteomics, inflammatory proteins, metabolomics).
• Developed and applied an Interpretable Multimodal Machine Learning (IMML) framework for predictive modeling of DSPN.
• Evaluated the predictive performance using Area Under the Receiver Operating Characteristic curve (AUROC).

Main Results:

• Clinical data alone achieved an AUROC of 0.752 for DSPN stratification.
• Predicting DSPN incidence over 6.5 years significantly improved (ΔAUROC > 0.1, AUROC 0.714) with clinical data plus two or more molecular modalities.
• Identified key predictive features including up-regulation of proinflammatory cytokines and down-regulation of SUMOylation pathways and essential fatty acids.

Conclusions:

• Identified potential novel biomarkers for incident DSPN, including specific molecular pathways and proteins.
• The findings suggest that IMML can guide prevention strategies for DSPN.
• Demonstrated the utility of the IMML framework for studying complex diseases with multimodal data.