On the left, the training data set sources and curation process are shown. The data type and the relevant information for the training process of each application are described in the horizontal arrows. On the right, the different neural network models built using the training data are represented. Application i uses protein ensembles as input and provides druggable conformations encompassing binding sites as output. Application ii is fed with the docked complexes between the druggable conformations and drug candidates as input, yielding hit compounds as output. Application iii takes the hit compounds as input to finally obtain the lead compounds.