Fig. 1

From the total dataset of amino acid sequences corresponding to different types of protein toxins with different modes of action in the cell (n = 27) and non-toxins (n = 1) randomly generated, the molecular descriptors PAAC and DPC were calculated. Subsequently, eight machine learning algorithms were evaluated, first on a training dataset (80%) which was subjected to tenfold cross-validation. Then, the generated models were evaluated on a test dataset (20%) (independent dataset). The final stage consisted of selecting the best predictive model for its incorporation into a web application called MultiToxPred 1.0