Fig. 2: Schematic of cTI application to detect disease-associated patterns and patient neuropathological stages in neurodegeneration¹⁶.

a In vivo blood (N = 744; ADNI) and post mortem brain (N = 1225; ROSMAP, HBTRC) tissues, from Alzheimer’s disease (AD), Huntington’s disease (HD), and/or normal controls (HC) subjects, are screened to measure the activity of ~40,000 transcripts. b Each population’s high-dimensional data are reduced to a set of disease-associated components via c contrastive principal component analysis (cPCA)^21,62. d This allows each subject to be represented in a reduced n-dimensional disease-associated space where the corresponding position reflects his/her pathological state (proximity to the bottom-left corner implies a pathology-free state; conversely, the top-right corner implies advanced pathology). For instance, when analyzing the GE data from the HBTRC’s highly heterogeneous population (including HC, LOAD, and HD subjects, total N = 736), the high-dimensional data were reduced to seven contrasted PCs [cPCs] capturing up to 97.5% of the population variance (and individually explaining 38.73%, 19.91%, 16.18%, 8.46%, 5.85%, 5.50%, and 2.87% of the variance, respectively). Notice that, for visualization simplicity, here were only represented the first three cPCs, but the quantitative analysis considers all identified components. Within this cPCs space, each subject is automatically assigned to a disease trajectory that represents a subpopulation of subjects potentially following a common disease variant (see “Methods”). The number of subpopulations (disease trajectories) is determined automatically based on how the subjects “cluster” together in the disease-associated space. e An individual molecular-disease score is then calculated, reflecting how advanced each subject is in his/her disease trajectory. This score significantly predicts neuropathological deterioration. f Finally, the resulting model weights (from contrastive PCA) allow the identification and posterior functional analysis of most influential genes/features. Panels a–c, e, f adapted with permission from ref. ¹⁶.