Machine Learning Models for the Classification of CK2 Natural Products Inhibitors with Molecular Fingerprint Descriptors

Abstract

Casein kinase 2 (CK2) is considered an important target for anti-cancer drugs. Given the structural diversity and broad spectrum of pharmaceutical activities of natural products, numerous studies have been performed to prove them as valuable sources of drugs. However, there has been little study relevant to identifying structural factors responsible for their inhibitory activity against CK2 with machine learning methods. In this study, classification studies were conducted on 115 natural products as CK2 inhibitors. Seven machine learning methods along with six molecular fingerprints were employed to develop qualitative classification models. The performances of all models were evaluated by cross-validation and test set. By taking predictive accuracy(CA), the area under receiver operating characteristic (AUC), and (MCC)as three performance indicators, the optimal models with high reliability and predictive ability were obtained, including the Extended Fingerprint-Logistic Regression model (CA = 0.859, AUC = 0.826, MCC = 0.520) for training test andPubChem fingerprint along with the artificial neural model (CA = 0.826, AUC = 0.933, MCC = 0.628) for test set. Meanwhile, the privileged substructures responsible for their inhibitory activity against CK2 were also identified through a combination of frequency analysis and information gain. The results are expected to provide useful information for the further utilization of natural products and the discovery of novel CK2 inhibitors.

MolToxPred: small molecule toxicity prediction using machine learning approach

“…Estate fingerprints (electropological), MACCS fingerprints (cover functional groups, ring systems), Morgan fingerprints (circular), Klekota-Roth fingerprint 91 (chemical substructures enriched for biological activity) and graph-based fingerprints are the crucial fingerprints in the top 50 features. Ext fingerprint is an extension of the CDK fingerprint, which takes into account the nature of the ring, including rich structural information 92 and is a part of the important feature list. These features' importance confirms that the structural and molecular properties of the compound are crucial for learning the relationship between molecules and their toxicity.…”

Deep learning based predictive modeling to screen natural compounds against TNF-alpha for the potential management of Rheumatoid Arthritis: Virtual screening to comprehensive in silico investigation

“…A negative logarithmic conversion was used to create a more uniformly distributed collection of IC 50 data, resulting in pIC 50 . The PaDEL software [23] was employed to generate PubChem fingerprints [24] from the canonical SMILES. PubChem fingerprints contained 881 binary descriptors which indicate the existence of specific chemical compound groups [25].…”

Implications of Additivity and Nonadditivity for Machine Learning and Deep Learning Models in Drug Design

“…Especially with recent advancement of deep learning, many methods have become available in order to predict molecular properties. State-of-the-art property prediction models make use of fingerprints as molecular representations. − Furthermore, models can be trained on SMILES representations or molecular graphs in order for the network to learn the important features themselves, without the need for precalculated molecular descriptors. − …”