Skip to main content
Fig. 3 | BMC Genetics

Fig. 3

From: Data mining and machine learning approaches for the integration of genome-wide association and methylation data: methodology and main conclusions from GAW20

Fig. 3

Deep learning model applied to genome-wide methylation data. Panel a represents an interconnected node (neuron), the basic element of artificial neural networks. an represents the nth input signal into the neuron; wn represents the corresponding weight of an; and b is a random bias added to avoid overfitting. The sum of multiplied input values and random bias z is transformed into an output value by a fixed activation function σ. Panel b shows the specific deep neural network model used to investigate GAW20 methylation data. The first layer (input layer) included all 463,995 CpG sites. The second and third hidden layers were configured to 500 and 250 nodes, respectively. The fourth layer (ReLu) aims to nonlinearity, and the fifth layer (Dropout) targets at overcoming overfitting

Back to article page
\