Figure 2

Schematic representation of a three-layer feed-forward neural network. Genotypes for 2421 (or 700) SNPs were used as inputs x _j  = {x _ij |i = 1, 2, …, n}, where n is the number of individuals with genotypes; each SNP was connected to up to 4 neurons via coefficients w _kj , where k denotes neuron and j denotes SNP; here, w _k is a weight from a hidden layer units to the output unit, f _k is an activation function applied to hidden layer units (e.g., the hyperbolic tangent), g is an activation function applied to the output layer unit (e.g., linear), b⁽¹⁾ and b² are biases of hidden and output layer units, and $\widehat{y}$ is a predicted value.