Machine learning-based approach to Predict Dual Inhibition Activity of Chemical Compounds on Lysine Specific Demethylase ۱ (LSD۱) and Histone Deacetylases (HDAC) for Cancer Treatment

Background and Aims: There are scientific evidences that Lysine specific demethylase ۱ (LSD۱) and Histone Deacetylases (HDAC) are functionally and physically related to each other in different cancers. Thus, the simultaneous inhibition of these targets will have synergistic effects for the treatment of different cancers. Until the present time, a lot of compounds have been reported as specific inhibitors of these two targets, separately. There are also some reports that have demonstrated the dual inhibition activity of some other compounds on LSD۱ and HDAC. Machine learning (ML) methodologies are being applied in many drug discovery procedures for prediction of biological activities of different compounds. In this study, different machine learning models have been trained based on the data related to the chemical structures of LSD۱ and HDAC inhibitors for prediction of dual inhibition activity of new molecules on these two receptors. Methods: The dataset was prepared by extracting the chemical structures (as SMILES notations) of compounds (in three groups) reported as dual inhibitors of LSD۱ and HDAC, LSD۱ inhibitors and HDAC inhibitors in different articles and also the ChEMBL database. The PubChem fingerprints descriptors were extracted as the independent features. And the IC۵۰s were considered as the target variable. The models that are used in this paper are Support Vector Machine (SVM), Random Forest and Logistic Regression. Further, to fine-tune hyper parameters using ۵-fold cross-validation, GridSearchCV was used and each model was trained and tested on the dataset. Results: The best ML model after hyper parameter tuning by GridSearchCV method was determined to be Logistic Regression based on the scores (۰.۹۸). Conclusion: In this study, Logistic Regression model was introduced as the best ML model for prediction of the dual inhibition activity of new molecules on LSD۱ and HDAC receptors. The compounds inhibiting simultaneously both receptors will be more effective than the compounds effecting just one target. Prediction of the fact that a new molecule will act as a LSD۱ or HDAC or a dual inhibitor, will be very helpful for reducing the total costs of the drug discovery procedures for the purpose of cancer treatment. Keywords: Artificial Intelligence, Machine

کلیدواژه ها:

Artificial Intelligence ، Machine