| Application: | 50063, Identifying genetic predictors of response to pharmaceutical drugs using machine learning |
| Title: | Polygenic modelling of treatment effect heterogeneity |
| Size: | 782812 B |
| Archived: | 29 Jul 2021 |
| Stability: | Complete |
| Personal: | No individual-level data |
Return 3693
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Notes
We developed an agnostic approach to combine genetic variants into a composite effect modifier (a polygenic response score) that divides the population into genetic subgroups which are predicted to respond differently to a particular treatment. As a proof of concept example, we leveraged the UK Biobank dataset to construct a polygenic response score for the effect of statins on LDL-cholesterol levels using random forest of interaction trees (RFIT). The clinical impact of this example was limited, where no more heterogeneity in the predicted treatment effect was detected than would be expected by chance alone. However, our approach may still be useful to detect individuals with particularly strong or weak predicted response to other particular treatments, leading to opportunities for precision medicine.Application 50063
Identifying genetic predictors of response to pharmaceutical drugs using machine learning
Large variability exists in each patient's response to pharmaceutical drugs, for example certain patients may benefit from taking a specific drug while other patients may not. The unique genetic profiles of each patient contain clues for identifying whether the patient is likely to respond or not respond to a specific drug. This project aims to leverage the large-scale genotype and phenotype data within the UK Biobank to construct a machine learning model which would generate predictions of response to a specific drug. As proof of concept, we will focus on statins, a class of drugs which lowers LDL cholesterol levels to reduce the incidence of cardiovascular diseases. The proposed method is both cost and time effective, where different from traditional approaches, it eliminates the need to conduct a separate clinical trial and genotype study for each drug of interest. Moreover, the proposed method could potentially be highly valuable in personalized medicine applications, which has a large public health impact. Specifically, the model could use the genetic information of each patient to inform doctors about a patient's magnitude of response to certain drugs, such that dosage could be adjusted to decrease side effects due to over-treatment. Moreover, the model could also inform doctors of non-responders, skipping the "trial and error" approach in the current standard of care for selecting an effective drug. The duration of the proposed project is 6 months.
| Lead investigator: | Mr Zhi Ming Xu |
| Lead institution: | University of Cambridge |
1 Publication
| Pub ID | Title | Author(s) | Year | Journal |
|---|---|---|---|---|
| 3694 | Polygenic modelling of treatment effect heterogeneity | Zhi Ming Xu (+1) | 2020 | Genetic Epidemiology |
Enabling scientific discoveries that improve human health