Trustworthy Enhanced Multi-view Multi-modal Alzheimer's Disease Prediction with Brain-wide Imaging Transcriptomics Data

TL;DR

This work tackles Alzheimer's disease prediction by integrating brain-wide transcriptomics with multi-modal imaging in a trustworthy multi-view framework. It constructs transcriptomics- and imaging-based ROI co-functional networks (T-RRI and R-RRIs), applies multi-level graph attention for view-specific representations, and fuses modalities through cross-view and cross-modal attention. A novel true-false-harmonized class probability (TFCP) mechanism calibrates modality confidence, enabling adaptive and reliable fusion across heterogeneous data. Experiments on AHBA+ADNI demonstrate superior predictive performance and yield biologically plausible ROI biomarkers, highlighting the practical impact of combining upstream transcriptomics with downstream imaging for disease diagnosis and interpretation.

Abstract

Brain transcriptomics provides insights into the molecular mechanisms by which the brain coordinates its functions and processes. However, existing multimodal methods for predicting Alzheimer's disease (AD) primarily rely on imaging and sometimes genetic data, often neglecting the transcriptomic basis of brain. Furthermore, while striving to integrate complementary information between modalities, most studies overlook the informativeness disparities between modalities. Here, we propose TMM, a trusted multiview multimodal graph attention framework for AD diagnosis, using extensive brain-wide transcriptomics and imaging data. First, we construct view-specific brain regional co-function networks (RRIs) from transcriptomics and multimodal radiomics data to incorporate interaction information from both biomolecular and imaging perspectives. Next, we apply graph attention (GAT) processing to each RRI network to produce graph embeddings and employ cross-modal attention to fuse transcriptomics-derived embedding with each imagingderived embedding. Finally, a novel true-false-harmonized class probability (TFCP) strategy is designed to assess and adaptively adjust the prediction confidence of each modality for AD diagnosis. We evaluate TMM using the AHBA database with brain-wide transcriptomics data and the ADNI database with three imaging modalities (AV45-PET, FDG-PET, and VBM-MRI). The results demonstrate the superiority of our method in identifying AD, EMCI, and LMCI compared to state-of-the-arts. Code and data are available at https://github.com/Yaolab-fantastic/TMM.

Figures (4)

• Figure 1: Framework overview. (a) Transcriptomics-specific RRI network (T-RRI) is constructed from the brain-wide gene expression data, and radiomics-specific RRI network is derived from each imaging data. (b) Three imaging modalities including VBM, FDG, and AV45 are employed in our analysis. (c) The TMM architecture. (i) For each sample, modality-specific T-RRI and R-RRI networks are constructed by mapping ROI measurements to each node. Multi-level GAT is applied to capture interactions within ROIs for each view, generating view-specific representations, followed by cross-view attention to create multi-view embeddings. (ii) The true-false-harmonized class probability (TFCP) is designed to estimate the prediction confidence for each modality. (iii) Cross-modal attention mechanisms fuse multiple modalities to deliver the final predictive output. (d-f) illustrates the details of cross-view graph fusion, the TFCP mechanism, and the cross-attention mechanisms, respectively. (g) The identified ROI biomarkers are annotated for their functional co-activation.
• Figure 2: (a) Performance comparison of different modality combinations on NC vs. AD. (b) Performance comparison of difference RRI thresholding hyperparameters on NC vs. AD.
• Figure 3: Transcriptomics-based (top panel, grey edges) and radiomics-based (bottom panel, green edges) connectivity of top ROIs derived from each modality.
• Figure 4: Coactivation maps of top ROIs obtained from NC vs. AD task.