Using Proteochemometrics models to predict interaction between isoforms of caspase and their inhibitors

In this study we will focus on proteochemometrics modeling which is a new computational approach to the study of drug design for prediction of the interactions between Caspase isoforms and their inhibitors. Caspase is a family of aspartate proteases, that play key roles in programmed cell death and inflammation but what is essential in apoptosis is the division of caspases into two groups of initiating caspases (caspase ۲, ۸. ۹ and ۱۰) and executive caspases (caspase ۳, ۶, and ۷) . Materials andMethods:In this project for modelling, we used protein and ligand descriptors. First, the ligands collected from Binding DB (in SDF format) were optimized with hyperchem, and ۱۴۴۴ ۱D and ۲D, along with ۴۳۱ ۳D ligand descriptors, were extracted with Padel software. Next, ۱۴۴۱ protein descriptors were extracted using the protr R package, yielding a total of ۳۳۱۷ descriptors. For feature selection we used, NCA (Neighborhood Component Analyses). For PCM modelling three models were developed such as SVR, Decision Tree, and Ensemble. Results andConclusion:A model induced by a machine learning algorithm should be validated. Common methods for validation are; application of the model to a test set, k-fold cross validation and randomized shuffing of the outcomes. The analysis based on R۲ and RMSE and Q۲ showed that the ensemble model was the best. R۲ in this model was ۰.۸۱, Q۲ was ۰.۷۶ and RMSE was ۰.۰۳۶. The criteria for an acceptable model R۲ was > ۰.۶ and Q۲ was > ۰.۵. The results of this study demonstrated that the Ensemble model has better performance than other models in terms of R۲, Q۲, and RMSE criteria