The BHF-Turing Cardiovascular Data Science Awards (Second Call): Molecular causal networks of hypertension – a machine learning approach (joint funding with The Alan Turing Institute)

Genome-wide association studies (GWAS) have found over a thousand genetic variants associated with blood pressure (BP). The biological underpinnings of their association with BP remain elusive.

We hypothesise that these variants operate through functional effects at various stages of the transcriptional regulation in the kidney (a key tissues of relevance to BP regulation).

Using a unique data resource (gene expression, methylation and splicing) collected from 500 human kidneys and published summary results of BP from large-scale GWAS (including the UK Biobank), we will develop a machine learning analytic method to explore the causal network underlying hypertension.

In particular, we will use random forest to select important predictive genetic variants; develop an undirected graph depicting correlations among the genetic and molecular variables on the basis of conditional independence properties, followed by inferring directionality of these correlations by the principle of Mendelian randomization.

The causal network built in this project will help gain relatively comprehensive insights into the biological mechanisms of hypertension and pinpoint novel diagnostic and therapeutic targets for cardiovascular medicine. This analytic framework can also be applied to causal network studies of a range of health conditions.