Multilayer Integration of Networks Toolbox (MINT)

We present MINT (Multilayer Integration of Networks Toolbox), a Python package for multimodal data integration and community detection. MINT includes data standardization, Similarity Network Fusion, Generalized Louvain clustering, visualization, cross-validation, and modality selection optimization, capturing complex relationships among disease markers. We applied MINT to two multimodal datasets spanning the Alzheimer's disease (AD) spectrum: a primary cohort of 206 participants and a validation cohort of 143 participants, including structural magnetic resonance imaging (MRI), amyloid positron emission tomography (PET), cerebrospinal fluid (CSF), cognition, and genetics. We hypothesized that modeling intra- and inter-modality associations would improve AD prediction and identify preclinical cases. Across both datasets, MINT identified PET and CSF as optimal modalities and detected two communities: one AD-dominant and one cognitively normal-dominant (CN). Sensitivity and specificity for CN and AD were 84.38% (95% CI: 73.14-92.24) and 92.65% (95% CI: 83.67-97.57). The AD-dominant community exhibited poorer cognition and higher genetic risk and AD pathology (p < 0.001). CN individuals in this group showed elevated amyloid (p=0.009), tau (p=0.004), and ptau (p < 0.001) compared to AD individuals in the CN-dominant group. MINT can identify biologically relevant subgroups, predict disease progression, and serves as a powerful tool for uncovering complex relationships across heterogeneous and multifactorial disorders.