Figure 4

eScience-Bayes applied to protein-protein interaction data. Modeling of the PDZ domain-peptide interaction data with the eScience-Bayes approach. (Color-coding is as in Fig. 1.) (A) All PDZ and peptide sequences and interaction data were retrieved from the supplementary material of Chen et al. [23] via PubMed Central. The PDZ sequences were aligned using the MUSCLE Web service. Prior information was retrieved using the SAS Web service. SAS was used to determine the number of contacts made between PDZ domain residues and ligand residues from the 3D complexes available in the Protein Data Bank. (B) The aligned PDZ domain and peptide sequences were characterized by numbers capturing their physicochemical properties. The characterizations of the PDZ domain-peptide interactions were correlated to a 1 or a 0 in a binary response variable, which indicated whether a PDZ domain and a peptide could bind or not to each other. Analogous to the breast cancer demonstration, we reduce the dimensionality of the model guided by the prior information derived in (A) (see Methods). (C) The models were fit on an HPC resource.