The research, led by Jiajing Hu and Alfredo Iacoangeli at King’s College London together with Ammar Al-Chalabi, Alexander Schönhuth and a large international network of ALS clinicians, builds on an earlier observation that Capsule Networks, or CapsNets, can learn to distinguish people with ALS from unaffected controls using nothing more than single nucleotide polymorphism genotyping data. CapsNets are a relative newcomer to the deep learning menagerie. Where conventional convolutional neural networks compress spatial information into a single scalar output at each layer, capsules preserve the pose and identity of features as vectors, allowing the network to reason explicitly about part-whole relationships. In genomics, that means the model can, in principle, capture how variants within a gene or pathway combine, rather than treating each single nucleotide polymorphism as an independent bit of evidence. The catch, until now, has been that published CapsNet experiments were confined to specific datasets and genotyping platforms, leaving open whether the approach would generalise beyond the population and the chip it was trained on.

The new work tackles that generalisation problem head-on with a dataset of unusual scale for this disease. The team assembled genetic data from more than 47,000 individuals across 13 countries, drawing on the Project MinE consortium and collaborating ALS centres in Europe, North America, the Middle East and Australia. Crucially, the samples were genotyped on nine different platforms, each of which interrogates a slightly different set of variants and produces its own batch effects, missingness patterns and allele-calling quirks. Any model that silently learns the fingerprint of a particular genotyping chip will fail the moment it meets a patient typed on another one, a failure mode that has plagued machine learning applications in medicine. The researchers therefore engineered their pipeline to be robust to these technical discrepancies, using quality control procedures, principal component analysis to characterise and adjust for ancestry, and a gene-based principal component strategy that compresses variant-level information into gene-level features before the network ever sees it.

That gene-based dimensionality reduction is one of the study’s key technical moves. Rather than feeding hundreds of thousands of raw SNP genotypes into the CapsNet, the authors aggregate variants by gene, computing principal components that summarise each individual’s genetic variation within each gene region. This does two things at once. It shrinks the input space to a size the network can learn from without overfitting, and it anchors the features in biological units, genes, that are shared across genotyping platforms even when the exact SNP content differs. The capsules then operate on these gene-level representations, and the routing-by-agreement mechanism of the CapsNet allows lower-level gene features to vote on the higher-level decision of whether the genome belongs to someone with ALS. The result is a model that is flexible enough to accept an individual external sample, whatever platform produced it, and still produce a calibrated risk estimate.

The performance results are striking for a disease whose common genetic architecture has proved so elusive. In external validation across diverse ALS populations from multiple countries, the model achieved high precision and sensitivity in separating individuals with ALS from unaffected controls, holding up despite the technical and ancestral heterogeneity of the data. The authors are careful to frame this as risk stratification and diagnostic support rather than a stand-alone diagnosis; the model does not replace clinical examination, electromyography or targeted mutation testing. But the ability to extract a disease signal from genome-wide genotyping data in sporadic ALS, where classical genetics has explained so little, suggests that the collective weight of thousands of common variants, and the nonlinear interactions among them, carries real information that linear polygenic risk scores have struggled to capture.

Perhaps the most provocative result comes from a simulation. The team modelled what would happen if the tool were deployed as a population screening test for ALS, and found that its predictive performance under that scenario was comparable to published estimates for screening based on the major known ALS-causing mutations such as FUS and C9orf72. That comparison matters because mutation-based screening only works for carriers of those specific variants, whereas the deep learning model applies to everyone. In effect, the study suggests that a genome-wide pattern learned by a neural network can match the yield of looking for the single most consequential known mutations, at least in the simulated screening context. If that holds up prospectively, it would mean genetic risk assessment could be extended to all individuals regardless of family history or mutation status, a genuine democratisation of ALS genetics.

The clinical logic is compelling because ALS is a disease where time is function. The average diagnostic delay from first symptoms to confirmed diagnosis remains long, often around a year, and there is growing evidence that the earlier treatment begins, the better the outcome. Riluzole, the oldest approved therapy, and the newer antisense and antibody-based approaches now entering the clinic all work best early. A validated genomic risk score could flag individuals at elevated risk for closer monitoring, help neurologists interpret ambiguous early presentations, and, once a diagnosis is made, contribute to prognostic stratification. The authors are explicit that their tool supports, rather than replaces, existing genetic testing: for the minority of patients with known pathogenic variants, targeted testing remains essential, particularly since those variants now determine eligibility for precision therapies such as antisense oligonucleotides.

The study also carries a sober set of caveats that the authors do not shy away from. The model was trained and validated on case-control data, which is the right starting point but not the same as predicting who will develop disease in a healthy population. ALS is rare, so even a highly precise classifier will generate false positives at population scale, and the ethical weight of telling a healthy person they carry elevated risk for a uniformly fatal disease is enormous. Ancestry is another live concern: despite the international scope of the dataset, most participants are of European descent, and the authors stress that further prospective clinical validation in diverse populations is required before any diagnostic use. The team’s own conclusion is measured: the method could support genetic risk stratification now, and diagnostic support in the future, but only after that prospective validation is done.

What makes the work resonate beyond ALS is its demonstration that deep learning can extract clinically meaningful signal from the genomic dark matter of a complex neurodegenerative disease. Capsule Networks were designed for computer vision, and their migration into genomics is part of a broader convergence in which architectures built for one domain prove unexpectedly adept at another. The gene-based principal component trick, the platform-agnostic preprocessing and the multi-country external validation form a template that other common-disease geneticists can copy. For the ALS community, which has watched genetic discoveries accumulate without translating into risk prediction for most patients, the study offers something rarer than a new gene: a working bridge between the genome-wide association era and the clinic. The 47,000 genomes behind the model were donated by patients and families across 13 countries, and the authors dedicate the tool’s promise to exactly that population, the sporadic cases who have, until now, been told their genetics held no answers.

Subject of Research: Deep learning genomic risk prediction for sporadic amyotrophic lateral sclerosis

Article Title: Towards a deep-learning genomic tool for risk stratification and diagnostic support in sporadic ALS

Article References: Hu, J., Pain, O., Al Khleifat, A., Shatunov, A., Andersen, P. M., Başak, N. A., Cooper-Knock, J., Corcia, P., Couratier, P., de Carvalho, M., Drory, V., Gotkine, M., Landers, J. E., Glass, J. D., McLaughlin, R., Pardina, J. S. M., Morrison, K. E., Pinto, S., Povedano, M., … Iacoangeli, A. (2026). Towards a deep-learning genomic tool for risk stratification and diagnostic support in sporadic ALS. Genome Medicine. https://doi.org/10.1186/s13073-026-01744-5

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01744-5

Keywords: ALS, deep learning, Capsule Networks, genomics, risk stratification, sporadic ALS, Project MinE, polygenic risk, genome-wide association study, diagnostic support, C9orf72, Genome Medicine

The apolipoprotein E gene, known universally as APOE, is the single most influential genetic risk locus for late-onset Alzheimer’s disease. It exists in three common forms: ε2, which appears to confer some protection against the disease; ε3, the most common allele; and ε4, which substantially raises risk in a dose-dependent manner, so that a single copy increases an individual’s lifetime risk several-fold and two copies raise it dramatically. Because ε4 has loomed so large in the genetics of Alzheimer’s, most genome-wide association studies have treated it as just another risk factor to be adjusted for statistically, entering it into models as a covariate rather than asking a more fundamental question: does the effect of every other risk variant in the genome depend on APOE status?

The research team behind the new study set out to answer precisely that question by stratifying their analyses. Rather than pooling all cases and controls into a single association test, they divided participants into groups based on APOE genotype—ε4 carriers and non-carriers—and performed separate genome-wide scans within each stratum. This design allowed them to detect variants whose association with Alzheimer’s disease is amplified in one genetic background and muted, or even abolished, in the other. It also enabled them to test explicitly for interactions between each variant and APOE status, a statistical approach known as genome-wide interaction analysis that demands very large sample sizes to achieve adequate power.

The results were striking. The analyses identified a set of variants whose effects on Alzheimer’s risk are attenuated by the ε4 allele—risk factors that matter substantially in people without ε4 but lose much of their influence among those who carry it. Other variants showed the opposite pattern, their effects augmented in the presence of ε4, suggesting that they operate along biological routes that converge with, or are magnified by, the pathways through which ε4 itself does damage. This bidirectional pattern of effect modification is exactly what one would expect if ε4 does not merely add risk in a straightforward additive fashion, but instead reshapes the disease process in ways that reorganize the contribution of the rest of the genome.

Technically, the study illustrates the power and the demands of stratified genome-wide association methodology. Standard GWAS meta-analyses pool tens of thousands of cases and controls and test each single-nucleotide polymorphism for an average effect across the entire sample. Stratified analyses effectively halve the available sample in each stratum, which means that detecting variants with modest effect sizes requires careful harmonization of datasets, consistent genotyping and imputation across cohorts, and statistical models that can borrow strength across strata while still testing for heterogeneity of effect. The researchers applied these methods across tens of thousands of Alzheimer’s disease cases and cognitively normal controls drawn from multiple cohorts, combining summary statistics within each APOE stratum and then formally comparing effect estimates between strata to flag loci exhibiting significant interaction with ε4 status.

The biological interpretation of the discovered loci points toward several well-known culprits in Alzheimer’s pathophysiology, but with new nuance. Variants near genes involved in cholesterol metabolism and lipid transport, for example, showed patterns of effect that differed markedly between ε4 carriers and non-carriers. This is biologically coherent, because the APOE protein is itself the principal carrier of cholesterol and other lipids in the brain, shuttling them between cells through receptors on neurons and glia. The ε4 allele encodes a protein that binds lipid particles poorly and is associated with impaired clearance of amyloid-beta, the peptide that accumulates into the plaques characteristic of the disease. If ε4 already compromises lipid handling and amyloid clearance, additional variants in the same pathways may be redundant in carriers while remaining consequential in non-carriers, where those pathways are still functioning closer to normal.

Other interacting loci highlighted genes tied to the innate immune system, particularly microglial pathways that have emerged as central players in Alzheimer’s biology. Microglia, the resident immune cells of the brain, clear debris and amyloid deposits, and risk variants in genes such as TREM2 and its network partners have previously been shown to alter microglial responses to pathology. The new stratified analysis suggests that the contribution of immune-related variants to disease risk is modulated by APOE status, reinforcing a picture in which lipid biology and neuroinflammation are not parallel, independent routes to dementia but interlocking processes whose interplay depends on which APOE allele a person inherits.

The clinical and research implications of this work are considerable. For genetic risk prediction, the findings caution against one-size-fits-all polygenic risk scores built from unstratified association data. A score calibrated on the full population may misestimate risk for ε4 carriers or non-carriers alike, because the weights it assigns to individual variants reflect averages that hold poorly in either subgroup. Stratified or interaction-aware polygenic models could sharpen risk stratification, which in turn matters for the design and interpretation of clinical trials: prevention studies that enroll participants by age and amyloid status may also need to account for how genetic background reshapes disease trajectories. Several ongoing anti-amyloid trials, for instance, have observed that treatment effects and side effects vary with APOE genotype, and the new results suggest that other genetic modifiers may warrant similar attention.

For basic researchers, the study offers a roadmap for dissecting mechanism. When a variant’s effect is contingent on APOE genotype, that contingency is itself a clue—a hint that the two genes’ products interact in the same cellular process, whether lipid transport, endosomal trafficking, immune activation, or amyloid metabolism. Functional follow-up in cell models such as induced pluripotent stem cell-derived microglia and neurons, engineered to carry different APOE alleles, could reveal how the identified variants produce their context-dependent effects. The stratified framework could also be extended to other forms of stratification, including sex, age at onset, or ancestry, each of which is known to modulate genetic risk in ways that pooled analyses obscure.

The study also carries a humbling message about the limits of aggregation. Decades of genome-wide association work have catalogued dozens of Alzheimer’s risk loci, but those catalogues describe average effects across mixed populations. The new results demonstrate that some of the genetic story has been hiding in the interactions—effects that cancel out or blur when ε4 carriers and non-carriers are analyzed together. As biobanks and consortium datasets continue to grow, interaction-aware analyses of this kind are likely to move from the margins to the mainstream of complex disease genetics, and Alzheimer’s disease, with its powerful APOE anchor, may prove to be the paradigm case. For the millions of families affected by dementia, the work underscores a shift already underway in the field: away from a single genetic explanation of Alzheimer’s and toward a more refined, personalized map of risk in which the answer to the question of what raises or lowers an individual’s chances increasingly depends on who that individual is, allele by allele.

Subject of Research: APOE-stratified genome-wide association analysis of Alzheimer's disease genetic risk

Article Title: APOE-stratified genome-wide association analyses provide insights into the genetic etiology of Alzheimers’s disease

Article References: APOE-stratified genome-wide association analyses provide insights into the genetic etiology of Alzheimers’s disease. (n.d.). https://doi.org/10.1038/s41588-026-02713-9

Image Credits: AI Generated

DOI: 10.1038/s41588-026-02713-9

Keywords: Alzheimer's disease, APOE, GWAS, genetics, ε4 allele, gene interaction, polygenic risk, microglia, lipid metabolism, amyloid-beta, APOE-stratified, genome-wide