Building Random Forest QSAR Models for Affinity Identification of 14-3-3 ζ with Optimized Parameters

14-3-3s present in multiple isoforms in human cells and mediate signal transduction by binding to phosphoserine-containing proteins. Previous studies demonstrate that the isoform 14-3-3 ζ acts as a key factor in promoting chemoresistance of cancer. Here, our work is devoted to developing the predictive models that can determine the binding affinity of phosphopeptide fragments against 14-3-3 ζ by the random forest approach. Based on the variable matrix built by the simple descriptors DPPS and statistical methods coupled with optimized hyperparameters, the robust models are obtained by a combinatorial peptide microarray dataset (n = 385 for N-terminal sublibrary, n = 384 for C-terminal sublibrary). For the test set, the R2 and RMSE are 0.8532 and 0.4516 at the N-terminal sublibrary (n = 96) and are 0.7998 and 0.5929 at the C-terminal sublibrary (n = 94), respectively. We also find that the distinct physiochemical properties function on the 14-3-3 ζ interaction. Overall, our results demonstrate that the computational methods based on QSAR analysis can be used for building the predictive models on the binding affinity of phosphopeptide against 14-3-3 ζ and contribute to the further research on clinical research.