GitHub - mew-two-github/predicting_pIC50_AKT: A Machine Learning model predicting pIC50 of certain drug molecules.

mew-two-github / predicting_pIC50_AKT Public

Notifications You must be signed in to change notification settings
Fork 1
Star 3

A Machine Learning model predicting pIC50 of certain drug molecules.

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 17 Commits
.ipynb_checkpoints		.ipynb_checkpoints
data		data
saved_models		saved_models
20200612- AKT model2.ipynb		20200612- AKT model2.ipynb
20200616 model2 correlation heat map.jpg		20200616 model2 correlation heat map.jpg
20200616- AKT Pred 12-06 final.ipynb		20200616- AKT Pred 12-06 final.ipynb
20200616- AKT model PCA.ipynb		20200616- AKT model PCA.ipynb
20200618- AKT_PaDEL- Autoencoder.ipynb		20200618- AKT_PaDEL- Autoencoder.ipynb
20200627- AKT model-2.ipynb		20200627- AKT model-2.ipynb
20200703- AKT Model3&Feature_selection.ipynb		20200703- AKT Model3&Feature_selection.ipynb
20200707- final_model.ipynb		20200707- final_model.ipynb
3DP.jpg		3DP.jpg
Exploring RDkit.ipynb		Exploring RDkit.ipynb
PDG- DPP-4 and AKT model2.ipynb		PDG- DPP-4 and AKT model2.ipynb
Padel.txt		Padel.txt
RDKit descriptors.ipynb		RDKit descriptors.ipynb
README.txt		README.txt
classification_scores.jpg		classification_scores.jpg
desc.csv		desc.csv
descriptors.csv		descriptors.csv
no_zeros_no_chembl.csv		no_zeros_no_chembl.csv
removed_zeros_AKT.csv		removed_zeros_AKT.csv
tSNE_3D.csv		tSNE_3D.csv
test.txt		test.txt
uneval_desc.csv		uneval_desc.csv

Repository files navigation

PREDICTIVE MODEL

This is an attempt to predict the pIC50 value of drugs which target AKT protein. The final model is an XGBoost model which has been deployed in the de novo design part.

All jupyter notebooks have been named according to what has been implemented in them.
Model 1: Based on PCA
Model 1.5: Based on Autoencoder
Model 2: Purely correlation based; also contains code used for deployment
Model 3: RandomForestRegressor models trained for various feature selection techniques and XGBoost is implemented after hyper-parameter tuning
final_model: The final model which has been deployed

no_zeros_no_chembl.csv: IMPORTANT FILE. It is the original dataset.(Padel descriptors for 2861 molecules)
descriptors.csv: used to store padel descriptors calculated in an experiment
uneval_desc.csv: used to store padel descriptors calculated in an experiment

SAVED_MODELS folder:
It contains the predictive models as well as associated data-processing files like scalers, pca methods, names of appropriate features and xml files for padel generation.
The final model is "best_from_gs38.model", final features are in "good_columns.pkl" and the appropriate XML file is "xg_desc3.xml"

DATA folder:
this contains several train-test-splits apart from outputs of feature selection techniques

MOL folder:
Contains molecules to be evaluated by PaDEL.