AI‐inferred gene expression trajectories mirror neuropathology and clinical deterioration in neurodegeneration

Abstract Background We lack robust minimally‐invasive molecular tests for early neurodegenerative detection, hindering the development of better treatment and therapies. Gene expression (GE) has been of crucial value for understanding neurodegenerative evolution, revealing disease‐specific differentiated genes/molecular‐pathways and networks. However, due to the large developing period of neurodegenerative disorders, we lack exhaustive longitudinal datasets covering the continuous molecular transitions underlying disease. Almost all our knowledge of the subjacent pathological mechanisms is based in data “snapshots” taken at a few disease stages. Method We collected in‐vivo GE samples from blood plasma of 744 subjects in the spectrum of late‐onset Alzheimer’s disease (LOAD) and from 1225 post‐mortem brains with LOAD or Huntington’s disease (HD), from ADNI, ROSMAP and HBTRC. The subjects have post‐mortem neuropathology evaluations (Braak, Cerad, Vonsattel staging) or amyloid/tau PET‐based quantifications. Next, we developed the GE contrastive Trajectory Inference (GE‐cTIF) algorithm, allowing the unsupervised machine‐learning identification of enriched GE temporal patterns in the diseased populations (e.g. LOAD) relative to a background population (healthy elderly). See Figure 1. Result Applied to the in‐vivo samples (Figure 2), the individual molecular pathological scores predicted tau positivity ( P <0.001, FEW‐corrected), amyloid positivity ( P <0.001, FEW), tau‐amyloid comorbidity ( P <0.001, FEW), clinical diagnosis ( P <0.001, FEW), memory performance ( P <0.001, FEW), executive function ( P <0.001, FEW) and future clinical conversion ( P <0.001). In the post‐mortem brains, the molecular scores were significantly predictive (P<0.001, FEW) of Braak, Cerad and Vonsattel stages. In addition, the proposed method allowed direct identification of genes and molecular pathways driving neurodegenerative progression. Notably, 85% and 90% of the highly predictive molecular pathways in the neurodegenerating brain (ROSMAP and HBTRC, respectively) were also among the most relevant pathways detected in the blood data (ADNI). Conclusion Results in three independent neurodegenerative datasets support the strong predictive power of GE‐cTI for predicting individual pathophysiology and cognitive decline. The obtained scores are direct measures of molecular integrity, calculated independently of phenotypic/clinical variables and able to be used as unbiased biomarkers in clinical applications. Our results have broad implications for uncovering dynamic mechanisms of molecular pathology, patient stratification in the clinic, and monitor response to personalized treatments.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.