FIGURE 1.

Classic PRSs and ML-driven PRSs the polygenic risk score for a target individual and phenotype of interest (y) is based on the individual's genetic data (x _g ) but can also include other data types (x _e ). The score is calculated using a linear regression (with weights β) or a machine learning model f _θ (e.g. a neural network with parameters θ). The parameters (β, θ) are learned using a separate training cohort. Note, however, that while the linear regession cofficients β are often publicly available or can be derived from published summary statistics, to train the neural network f _θ it is necessary to have access to individual level data in the training cohort.