From Association to Cause: Review Maps Causal Inference Methods for Alzheimer’s Molecular Data

Genome-wide studies have uncovered a vast catalogue of molecular associations with Alzheimer’s disease, but a new open-access review argues that association alone cannot reveal which changes actually drive the disease, which are mere consequences, and which might point to treatable targets. The review, published in Alzheimer’s Research & Therapy on 17 October 2026 by Rebecca Ting Jiin Loo and Enrico Glaab, surveys the methods researchers use to infer causal relationships from Alzheimer’s molecular data and assesses what those methods have found so far.

The authors examined causal-inference work spanning five layers of molecular information: genetics, transcriptomics, epigenomics, proteomics and metabolomics. The toolbox they describe is broad. Mendelian randomization uses genetic variants as instruments to test whether an exposure genuinely influences disease risk. Probabilistic graphical and Bayesian network models reconstruct likely directional relationships among molecules. Causal reasoning applied to knowledge graphs leverages curated biological relationships, while time-resolved approaches such as Granger causality exploit temporal ordering in dynamic data. Mediation analysis, meanwhile, probes whether an observed effect travels through an intermediate molecule or pathway.

According to the review, genetic-instrument and network methods have delivered concrete results: they have identified candidate causal molecules, mapped disease-relevant pathways and lent support to some existing drug targets. Yet the picture is far from settled. Several associations repeatedly reported in observational studies have not been supported when tested with genetic analyses, a discrepancy the authors say matters for anyone hoping to translate molecular findings into therapies.

One of the review’s central contributions is its explanation of why studies appear to disagree. Apparent conflicts, the authors argue, often reflect differences in the estimand — the specific quantity each analysis is trying to estimate — or in the tissue from which genetic instruments are derived, since instruments built from brain-relevant data may answer different questions than those built from blood or other peripheral tissues. Differences in the assumptions underlying each method can also produce divergent conclusions from the same underlying data.

The review closes with a roadmap for moving the field forward. The authors call for brain-relevant genetic instruments, longitudinal multi-omics datasets that track molecular changes over time, single-cell resolution to capture cell-type-specific effects, and triangulation across multiple causal-inference methods. Only such combined evidence, they conclude, can reliably separate cause from consequence in Alzheimer’s disease and prioritize intervention targets for experimental validation — a necessary step before candidate causal molecules can become validated drug targets.

Subject of Research: Causal inference methods applied to Alzheimer's disease molecular data

Article Title: Causal analysis of Alzheimer’s disease molecular data

Article References: Loo, R. T. J., & Glaab, E. (2026). Causal analysis of Alzheimer’s disease molecular data. Alzheimer's Research & Therapy. https://doi.org/10.1186/s13195-026-02235-w

Image Credits: AI Generated

DOI: 10.1186/s13195-026-02235-w

Keywords: Alzheimer's disease, causal inference, Mendelian randomization, Bayesian networks, multi-omics, genetic instruments, transcriptomics, proteomics, metabolomics, Granger causality, drug targets, review