Skip to content

Subtype-Prediction Polygenic Risk Scores

Summary

The subtype-prediction approach takes partitioned polygenic scores for already-recognized clinical or molecular subtypes as given — typically reusing clusters derived by a mechanism-first analysis — and asks a narrower question: do these fixed subtype-specific scores predict clinically meaningful outcomes when applied in new, large, independent cohorts? This is a validation/application step rather than a discovery step, and it is where partitioned PRS most directly demonstrate clinical utility.

T2D Partitioned PRS Applied Across 13 Cohorts: DiCorpo et al. 2022

DiCorpo, LeClair, Cole et al. took five T2D pathway-based partitioned PRS from Udler et al.'s clustering (β-cell, proinsulin, obesity, lipodystrophy, liver/lipid — see Mechanism-First Polygenic Risk Score Clustering) and tested their association with clinical outcomes across 13 cohorts from the CHARGE Consortium plus UK Biobank (AGES, ARIC, BioMe, Doetinchem, FHS, GENOA, Health ABC, MGB, MESA, NEO, PROSPER, Rotterdam, UKBB), totaling 454,193 participants (25,015 with T2D).[1] Outcomes tested included coronary artery disease (CAD), eGFR-creatinine, chronic kidney disease, systolic/diastolic blood pressure, hypertension, insulin use, and age at T2D diagnosis.[1]

Bonferroni-significant associations included:

Partitioned PRS Outcome Effect
Lipodystrophy Hypertension OR = 1.20, P = 6.8×10⁻¹⁵
Obesity Hypertension OR = 1.07, P = 1.3×10⁻¹¹
Liver/lipid CAD OR = 0.92, P = 1.2×10⁻⁴
Liver/lipid eGFR β = −0.81 mL/min, P = 9.4×10⁻⁶¹

These results show each pathway-derived score carries a distinct, outcome-specific signature at the largest scale tested to date for T2D partitioned PRS — directly supporting the clinical premise of mechanism-first subtyping (that the clusters are not just statistically distinct but differentially predictive of real complications) without deriving any new clustering.[1]

Relationship to Other Approaches

Subtype-prediction studies are often the natural third step after a mechanism-first discovery paper: Udler et al. (2018) and Kim et al. (2023) derived the T2D clusters; DiCorpo et al. (2022) applied them at far larger scale to test clinical-outcome prediction. Li et al.'s prediabetes work (Individual-First Polygenic Risk Score Clustering) sits between the two: it reuses Udler's five partitioned scores as fixed inputs (subtype-prediction-style provenance) but then re-clusters people by their profile across them (an individual-first analytical step) rather than simply regressing outcomes on the five scores directly.

See Also

Citations

[1] DiCorpo, D., LeClair, J., Cole, J. B., Sarnowski, C., Ahmadizar, F., Bielak, L. F., Blokstra, A., Bottinger, E. P., Chaker, L., Chen, Y.-D. I., Chen, Y., de Vries, P. S., Faquih, T., Ghanbari, M., Gudmundsdottir, V., Guo, X., Hasbani, N. R., Ibi, D., Ikram, M. A., Kavousi, M., Leonard, H. L., Leong, A., Mercader, J. M., Morrison, A. C., Nadkarni, G. N., Nalls, M. A., Noordam, R., Preuss, M., Smith, J. A., Trompet, S., Vissink, P., Yao, J., Zhao, W., Boerwinkle, E., Goodarzi, M. O., Gudnason, V., Jukema, J. W., Kardia, S. L. R., Loos, R. J. F., Liu, C.-T., Manning, A. K., Mook-Kanamori, D., Pankow, J. S., Picavet, H. S. J., Sattar, N., Simonsick, E. M., Verschuren, W. M. M., Willems van Dijk, K., Florez, J. C., Rotter, J. I., Meigs, J. B., Dupuis, J., & Udler, M. S. (2022). Type 2 Diabetes Partitioned Polygenic Scores Associate With Disease Outcomes in 454,193 Individuals Across 13 Cohorts. Diabetes Care, 45(3), 674–683. DOI: 10.2337/dc21-1395. Source: dicorpo2022-t2d-partitioned-prs-outcomes.md. Supports: all cohort, sample-size, and outcome-association results above. Location: Full text — Methods and Results.