Fig. 6

GPCR generic position mapping of most important rs3DDPD features in PCM regression tasks. a Top 25 most important features in PCM regression models using a temporal split validation for the GPCR set. The importance was averaged across the ten random seeds trained and the SD represented as error bars. Rs3DDPD features are mapped to their corresponding GPCR Ballesteros-Weinstein number or, if not available, region of the protein. b Representation of the most important rs3DDPD features in regression temporal split in the adenosine A1 receptor (PDB 5UEN). c Top 25 most important features in PCM regression models using a random split validation. d Representation of the most important rs3DDPD features in regression random split in the adenosine A1 receptor)