Mutual learning for joint disease detection and severity prediction reveals multimodal pathogenesis for neurodegenerative disorders

Abstract

Motivation: Neurodegenerative disorders influence millions of people worldwide, and uncovering the pathogenesis is of urgent need. Many efforts have been made to detect or predict neurodegenerative disorders, while exploring the pathogenesis has been ignored from a systemic perspective.

Results: To handle this issue, we propose a novel and powerful method, referred to as Pathogenesis-aware Mutual-Assistance Classification and Regression Optimization (Pa-MACRO). First, Pa-MACRO incorporates a mutual-assistance bidirectional mapping technique with a joint-embedding fine-grained interpretability module. This can extract the intrinsic factors and their interactions of multimodal pathogenesis. Second, our method can simultaneously classify an at-risk individual and predict the severity triggered by neurodegenerative disorders. Furthermore, to address the small sample size issue and the high-dimensional issue, we meticulously incorporate a semi-supervised cooperative learning method to integrate unlabeled data and extend it to a chromosome-wide setting in the spirit of divide-and-conquer. The Alzheimer's Disease Neuroimaging Initiative (ADNI) database was used to evaluate Pa-MACRO. Without bells and whistles, Pa-MACRO establishes new state-of-the-art results in various settings while maintaining superior interpretability, verifying its power and versatility in revealing the pathogenesis of neurodegenerative disorders.

Availability and implementation: The software is publicly available at https://github.com/ZJ-Techie/Pa-MACRO.

Figure 1.

A schematic illustration of a mutual learning-assisted predictive architecture with interpretable multimodal pathogenesis embeddings. First, Pa-MACRO incorporates a mutual-assistance bidirectional mapping structure with joint-embedding fine-grained interpretability, which helps collaboratively extract the intrinsic unity and interactions of multimodal pathogenesis that contribute to imaging phenotypes. Second, we meticulously design a mutual-assistance joint classification/regression module to ensure joint detection/assessment, which promotes disease detection and severity prediction.

Figure 2.

Comparison of evaluation outcomes for AD diagnosis (ACC, F1) and assessment (CCC) on VBM (a, b), where GE denotes inclusion of gene–environment interactions. Error bars indicate the standard deviation (SD) across five-fold cross-validation runs.

Figure 3.

Average canonical weights of SNPs derived from five-fold cross-validation. Each row corresponds to a method: (i) SMCCA; (ii) AdaSMCCA; (iii) RelPMDCCA; and (iv) Proposed.

Figure 4.

Average canonical coefficients of proteomic markers. Each row represents a method: (i) SMCCA; (ii) AdaSMCCA; (iii) RelPMDCCA; and (iv) Proposed.

Figure 5.

(a) Visualization of the detected brain imaging quantitative traits (QTs). (b) Heatmap depicting pairwise correlations between the top 10 SNPs and biomarkers, where the symbol “$\times$” denotes statistically significant associations (P < .05).

Figure 6.

The PheWAS investigation produced findings for SNP rs4420638 (APOC1).

Figure 7.

Correlation between the top-selected VBM imaging quantitative traits (QTs) and the clinical cognitive score RAVLT. Subfigures (a–f) respectively illustrate the associations between RAVLT and the regional gray matter volumes of: (a) Right Angular Gyrus, (b) Left Temporal Pole, (c) Right Temporal Pole, (d) Right Temporal Middle Gyrus, (e) Left Hippocampus, and (f) Right Hippocampus.

Figure 8.

Analysis of the mediating role of endophenotypic traits on diagnostic outcomes for top-selected genetic variants: (a) Diagnostic status, (b) ADAS, (c) MMSE, and (d) RAVLT. *P < .05, **P < .005, ***P < .001.

Figure 9.

Comparison of testing results for AD diagnosis (ACC, F1) and assessment (CCC) on FreeSurfer (a, b) datasets.