Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamic causal discovery in Alzheimer's disease through latent pseudotime modelling

Natalia Glazman, Jyoti Mangal, Pedro Borges, Sebastien Ourselin, M. Jorge Cardoso

TL;DR

The paper tackles the challenge of static causal graphs in Alzheimer's disease by introducing a latent pseudotime modelling framework to capture disease progression. It employs Bayesian Networks with Latent Time Embedding (BN-LTE) on ADNI data with 16 variables to learn dynamic causal graphs along pseudotime, incorporating disease-agnostic background knowledge to stabilize inference. The approach yields diagnostic performance of AUC $=0.82$ using pseudotime (compared to $0.59$ with age) and recovers known pathways such as NfL driving hippocampal atrophy and pTau elevations increasing NfL, while revealing dynamic interactions among emerging biomarkers like GFAP. This framework has translational potential for trial design and precision medicine, though it requires longitudinal validation and handling of unobserved confounders in larger, diverse cohorts.

Abstract

The application of causal discovery to diseases like Alzheimer's (AD) is limited by the static graph assumptions of most methods; such models cannot account for an evolving pathophysiology, modulated by a latent disease pseudotime. We propose to apply an existing latent variable model to real-world AD data, inferring a pseudotime that orders patients along a data-driven disease trajectory independent of chronological age, then learning how causal relationships evolve. Pseudotime outperformed age in predicting diagnosis (AUC 0.82 vs 0.59). Incorporating minimal, disease-agnostic background knowledge substantially improved graph accuracy and orientation. Our framework reveals dynamic interactions between novel (NfL, GFAP) and established AD markers, enabling practical causal discovery despite violated assumptions.

Dynamic causal discovery in Alzheimer's disease through latent pseudotime modelling

TL;DR

The paper tackles the challenge of static causal graphs in Alzheimer's disease by introducing a latent pseudotime modelling framework to capture disease progression. It employs Bayesian Networks with Latent Time Embedding (BN-LTE) on ADNI data with 16 variables to learn dynamic causal graphs along pseudotime, incorporating disease-agnostic background knowledge to stabilize inference. The approach yields diagnostic performance of AUC using pseudotime (compared to with age) and recovers known pathways such as NfL driving hippocampal atrophy and pTau elevations increasing NfL, while revealing dynamic interactions among emerging biomarkers like GFAP. This framework has translational potential for trial design and precision medicine, though it requires longitudinal validation and handling of unobserved confounders in larger, diverse cohorts.

Abstract

The application of causal discovery to diseases like Alzheimer's (AD) is limited by the static graph assumptions of most methods; such models cannot account for an evolving pathophysiology, modulated by a latent disease pseudotime. We propose to apply an existing latent variable model to real-world AD data, inferring a pseudotime that orders patients along a data-driven disease trajectory independent of chronological age, then learning how causal relationships evolve. Pseudotime outperformed age in predicting diagnosis (AUC 0.82 vs 0.59). Incorporating minimal, disease-agnostic background knowledge substantially improved graph accuracy and orientation. Our framework reveals dynamic interactions between novel (NfL, GFAP) and established AD markers, enabling practical causal discovery despite violated assumptions.

Paper Structure

This paper contains 8 sections, 2 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Results
  4. Discussion
  5. Acknowledgments
  6. Supplementary material
  7. MCMC procedure and convergence statistics
  8. Consensus and Estimated Causal Graph

Figures (3)

  • Figure 1: Demographic variables and biomarkers plotted as a function of the posterior mean of the inferred pseudotime values (z) of a single chain inferred by the BN-LTE model, with patients colour-coded according to their diagnosis.
  • Figure 2: Established consensus graph compared to estimated causal graphs presented as adjacency matrices with edge indices averaged across pseudotime and over the four chains, resulting in an edge inclusion probability (only PIP > 0.5 shown). The matrices on the diagonal reflect the resulting matrices from different settings, while the strictly lower triangular matrices reflect differences between each pair of matrices.
  • Figure 3: Biomarkers plotted against pseudotime, with solid lines representing the mean baseline trajectory of biomarkers and the corresponding effect of changing a specific parent variable. Parent variables are changed to the 5th, 50th, and 95th percentiles of the standardized values. The solid lines correspond to the mean of posteriors with shaded areas being 95% CI.