Double Machine Learning for Block‐Missing Data: Integrative Analysis of Plasma Biomarkers and Hippocampal Atrophy in Alzheimer's Disease Progression

Yue Yang, Ting Li, Borui Tang, Tengfei Li, Hongtu ZhuAlzheimer 2025 · DOI

TL;DR

This paper introduces a novel "double machine learning" framework to tackle the challenge of missing data in complex, multi-modal Alzheimer's disease (AD) datasets. By integrating plasma biomarker data with neuroimaging features, they identified key indicators of AD progression, offering a more robust understanding of the disease and potential avenues for early detection and targeted treatments.

What Problem Does This Paper Address?

Alzheimer's disease (AD) is a devastating neurodegenerative disorder characterized by progressive cognitive decline. Early detection and understanding of its progression are crucial for developing effective interventions, yet this is hampered by the complexity of AD's underlying mechanisms and the challenges in analyzing the vast amounts of data generated from different sources. Specifically, datasets often suffer from "block-missing data," where entire modalities or features are absent for certain individuals. This makes it difficult to perform integrative analyses that leverage the complementary information from various data types, such as plasma biomarkers and brain imaging. Traditional methods struggle to handle such missingness, potentially leading to biased results and an incomplete picture of the disease.

Methods

The core innovation of this paper is a novel double machine learning (DML) framework designed to address block-missing data in multi-omics analyses. While the paper doesn't detail the specifics of the "double machine learning" algorithm itself, it emphasizes its ability to de-bias the analysis. This suggests a two-stage approach: first, using machine learning to model the relationships within the available data for each modality, and second, using these models to estimate and impute missing values or to adjust for confounding factors introduced by missingness. This approach is combined with rigorous statistical analysis and state-of-the-art machine learning techniques to ensure robust and reproducible findings. The goal is to enable a comprehensive evaluation of causal relationships between plasma biomarkers and hippocampal atrophy in the context of AD progression.

Datasets

The paper describes an integrative analysis of plasma biomarkers (specifically mentioning GFAP, NEFL, GDF15, and LTBP2) and hippocampal atrophy (derived from neuroimaging). The neuroimaging data includes measures of functional connectivity, white matter integrity from diffusion MRI, and hippocampal subfield volumes. The paper does not explicitly name the source of the dataset or mention the creation of a new benchmark.

Key Insights and Ideas

  • The Power of Integration: Combining plasma biomarkers with neuroimaging provides a more comprehensive view of AD pathology than either modality alone.
  • Beyond the Obvious: The study identifies not only well-known AD biomarkers but also lesser-known proteins and neuroimaging features that are significantly associated with disease progression.
  • Unraveling Complex Pathways: The identified biomarkers point to specific biological processes involved in AD, such as astroglial activation, axonal injury, inflammation, metabolic dysregulation, and immune responses.
  • Hippocampus Remains Central: The detailed analysis of hippocampal subfields reinforces its critical role in AD pathology.

Key Results

The study identified several significant protein biomarkers and neuroimaging features associated with AD. The most prominent protein biomarkers were GFAP, NEFL, and CST5, implicated in astroglial activation, axonal injury, and inflammatory pathways, respectively. Functional connectivity disruptions were observed, along with alterations in metabolic and immune-related mechanisms indicated by other proteins. White matter integrity measures revealed microstructural damage in tracts crucial for cognitive and memory functions. Analysis of hippocampal subfields further highlighted the hippocampus's central role in AD pathology.

Limitations and Future Work

The paper does not explicitly detail limitations or future work within the provided text. However, the development of a novel DML framework suggests that overcoming data missingness in multi-modal AD research is an ongoing challenge. Future work would likely involve validating these findings in larger, independent cohorts and exploring the clinical utility of these identified biomarkers for early diagnosis and treatment monitoring. Further refinement of the DML framework to handle different types of missing data patterns could also be a direction.

Summary updated

Abstract

By providing a comprehensive analysis of the key biomarkers and neuroanatomical changes, our study advances the understanding of Alzheimer's disease progression and opens avenues for the development of targeted treatments.
BibTeX
@inproceedings{yang2025double,
  author = {Yue Yang and Ting Li and Borui Tang and Tengfei Li and Hongtu Zhu},
  title = {{Double Machine Learning for Block‐Missing Data: Integrative Analysis of Plasma Biomarkers and Hippocampal Atrophy in Alzheimer's Disease Progression}},
  booktitle = {Alzheimer},
  year = {2025},
  doi = {10.1002/alz70856\_104028},
}
Figure