To perform global sensitivity analysis One needs to save the model function of a trained model which is trained using "Regression Learner". The function name is trainRegressionModel. The C-code modifies the function "trainRegressionModel" by omitting a specific descriptor. Then one do "nrun" training to investigate the indispensability casued by the variable group.
gsa.m: sample matlab code modify_fun_matlab.exe: function to remove feature groups, to be called by gsa.m modify_fun_matlab.c: C-code.
===========================================================================
Part 2. Predict if a set of compounds are aggregators or not using two established machine learning models
Step 1A is to generate FP2 fingerprint using openbabel, and then run hex2bin to get binary file.
babel -imol2 sample.mol2 -ofps sample_FP2.fs -xfFP2 hex2bin.exe -i sample_FP2.fs -o sample_FP2.txt
Step 1B is to generate rdkit descriptor using python code. More details please refer to https://www.rdkit.org/docs/index.html. The rdkit properties are listed in 'rdkit_prop.list'.
Step 2. make prediction using Matlab. load SVM_fp2.mat or SVM_rdkit.mat first, then run SVM_fp2.predictFcn or SVM_rdkit.predictFcn to make prediction.
Alternatively, you may make prediction using the provided programs (predict_fp2.exe or predict_rdkit.exe) compiled in Windows:
predict_rdkit.exe -i test_rdkit.txt -o predict_rdkit.out
predict_fp2.exe -i test_fp2.txt -o predict_fp2.out