For example, Leu51, whose side chain forms part of a small hydrophobic cluster with the methyl group of Thr195 and the Pro257 side chain directly beneath the protein surface, is sensitive to hydrophobicity favoring more hydrophobic residues. Position 257 responds to log x Hydrophobicity, which means that amino acids simultaneously hydrophobic and soluble are favored. No property was selected for Thr195, due to the small dispersion of ����Gstat values at that position and its high tolerance to substitution. These are examples of constraints that act to stabilize superficial hydrophobic clusters, as opposed to those clusters truly at the protein core which are VE-822 better described by dependencies on amino acid volume, steric hindrance and FoldX predictions. Another example is the solvent-exposed Gln205 which exhibits a dependence on hydrophobicity contrary to that of Leu51. Gln205 is selected for its hydrophilicity, probably to aid in solubilizing the protein. Notably in this case, like in a few others, the ����Gstat values distribute in a cluster of high probability and another of low probability. Accordingly, hydrophobicity acts as a discrete classifier rather than as a variable for continuous modelling. Asn170, where hydrophilic residues are also favored, provides a more interesting example. This residue is on the protein surface but just 20.1% of its surface is exposed, thus it is not clear whether it contributes to solubility. However, Asn170 forms a hydrogen bond to Glu166, an active site residue, and could thus play an auxiliary catalytic role modulating the pKa and/or orientation of its LY2157299 carboxylate group. This is expected to be an important constraint, and indeed, the native amino acid is an outlier itself in the plot against hydrophobicity, being much more preferred than any other amino acid. This offset is observed in many correlations, suggesting that the wild type residues are largely preferred due to very specific reasons at some locations, although they can be substituted under certain constraints. In this example, Asn170 might be replaced by other residues that will not form exactly the same hydrogen bond with Glu166 but will at least preserve the polarity of the region. Notice that the composite variable Hydrophobicity x Flexibility explains slightly better the ����Gstat distribution for Asn170, which could account for a secondary need for rigidity to better position the side chains for hydrogen bond formation. The next two cases are examples of unsuspected relationships that this analysis helped unveil. First, Glu281 correlates positively with Volume/P and negatively with P /P.