Fig. 1

For the i-th side effect, the NNPS architecture is used. a Concatenation of the PCA representation of mono side effects \((F_{1})\) (blue) and the PCA representation of drug–protein interactions \((F_{2})\) (green). b Sum of the i-th and j-th rows in the drug features matrix for each \(d_{i}\) and \(d_{j}\) drug pair. c A three-layer neural network that computes the probability \(p_{i}\) and classifies the i-th side effect based on the threshold i