Mendelnoulli tracks how inherited risk changes across the life course
Inherited variation is fixed at birth, but its clinical effect is not static. This preprint introduces Mendelnoulli, a longitudinal model built to capture how both common polygenic risk and rare monogenic disease predisposition evolve with age and shape the sequence of diagnoses that accumulate over time. The premise is clinically intuitive: two people can carry genetic risk for the same disorder, yet one may present early, another late, and a third may never cross the threshold for disease within the observed follow-up window. Mendelnoulli was designed to separate overall disease burden from the specific diseases an individual is prone to develop, while also allowing polygenic and monogenic influences to act differently across latent disease trajectories.
The model was applied across 348 diseases, 36 polygenic scores, and 34 monogenic conditions in three biobanks. Those are broad inputs, and they matter because the central question is not whether genes influence disease, but when and how those influences become visible in real-world records. A static risk score treats age as a backdrop. Mendelnoulli treats age as part of the signal.
Age shifts the predictive value of common and rare variation
The clearest finding is that polygenic effects changed significantly with age. In held-out one-year prediction, Mendelnoulli outperformed a comparator model in which polygenic effects were assumed to remain constant over time for all 28 diseases tested. Median discrimination improved from an AUC of 0.92 to 0.95. In a field where small changes in discrimination can alter how many patients are flagged for follow-up, that difference is clinically meaningful. It also reflects biology: common variant burden does not simply add risk in a fixed way; it can interact with developmental stage, cumulative exposure, and the age at which organ systems become vulnerable.
One sentence captures the operational consequence: for individuals with no prior diagnoses, the model identified 8.6 more of every 100 new cases when each model flagged the highest-risk 5% of participants. In practical terms, that means earlier recognition of future disease burden without requiring a new biomarker assay or a novel sequencing platform. The information is already embedded in the genome; the model is attempting to read it with more temporal precision.
Rare variants followed disease timing that matched known biology
Mendelnoulli also separated signals from monogenic conditions and found that, for genes with an established disease association, the model selected that disease for 27 of 30 genes and ranked it first for 17. Seven would have been expected by chance. This is a strong face-validity result for a model that claims to disentangle which phenotype belongs to which inherited lesion, especially in a longitudinal setting where some people carry multiple genetic liabilities and only some liabilities will declare themselves during the available follow-up.
The model also recovered known differences in timing. A notable example was hereditary breast cancer: BRCA1-related effects peaked earlier than BRCA2-related effects, at 48 years versus 66 years. That age separation is not a trivial modeling artifact. It mirrors longstanding clinical observations that penetrance is not only a question of whether a variant is pathogenic, but also when it becomes pathogenic in a way that changes screening, counseling, and surgical decision-making. A framework that can reproduce this kind of age-specific divergence is closer to bedside reality than a single lifetime risk estimate.
Why a unified trajectory model could matter in practice
The biological logic behind the approach is straightforward. Common polygenic burden likely influences disease propensity through many small effects spread across pathways, while rare variants often exert stronger, more disease-specific effects. When those influences are collapsed into a single lifetime score, the resulting estimate can miss the age window in which risk is most actionable. A longitudinal model can instead describe whether a person is moving toward broad multimorbidity or toward a narrower set of genetically favored conditions. That distinction affects how risk is communicated and how surveillance is staged over time.
Mendelnoulli is most interesting where common and rare variation overlap. A patient may carry a polygenic burden that nudges cardiometabolic disease earlier in adult life while also harboring a high-impact variant that concentrates risk in a specific organ system. Traditional tools are often built to answer one of those questions at a time. This model attempts to answer both simultaneously, and to do so on a clock rather than in a vacuum.
The cohort breadth is a strength, but preprint status limits clinical translation
The analysis drew on three biobanks and used linked health records, which gives the work real-world depth. It also comes with the usual constraints of secondary data: ascertainment depends on the structure of the underlying record systems, disease coding varies across sites, and performance in biobank volunteers does not automatically carry into other populations or clinical settings. The preprint also remains uncertified by peer review, so the results should be treated as an advance in modeling rather than a practice-changing tool.
Even so, the study points to a concrete next step for genomic medicine. If age modifies genetic risk, then risk models should be built to let age modify genetic risk. Static scores may still have utility, but they are likely leaving information on the table, especially when the clinical question is not simply whether a person is at risk, but when that risk is most likely to become visible in the chart.