AI-assisted Quantitative Structural Activity Relationship (QSAR) Analysis

For a hit with little structural information, its biological effects can be predicted using other similar compounds' data, which is due to the similar structurally having similar physical and biological properties. In recent decades, significant advances have been made in the development of computer models for predicting various biological and chemical activities that help screen potential compounds with robust properties. Among the major approaches, quantitative structure-activity relationship (QSAR) modeling emerges as a valuable tool for the prediction of complicated physicochemical/biological properties of chemicals from their simpler experimental or calculated properties. Using QSAR, the researcher may establish a reliable quantitative relationship between structure and activity that will be utilized to derive an in silico model that will predict the activity of new compounds before they are synthesized. Classical QSAR studies include ligands with their inhibition constants, rate constants, binding sites, and other biological endpoints, as well as ligands with molecular properties such as polarizability, electronic, lipophilicity, and steric properties or with certain structural features. Creative Biolabs has established the AI-assisted bioinformatics analysis platform to support our customers in molecular docking, molecular dynamics simulation, virtual screening, and quantitative structural activity relationship (QSAR) analysis.

Fig.1 Schematic representation of the QSAR modeling workflow. (Nantasenamat, 2020)

2D-QSAR

In general, the molecular structure of a compound contains information that determines its physicochemical properties, which in turn determine the biological activity of the compound. The QSAR models describe the relationship between molecular structure and biological activity. The 2D-QSAR method is a model of the link between the chemical structure and the physiological activity for drug design, which employs the overall structural properties of the molecule as a parameter to perform regression analysis on the physiological activity of the molecule. Free-Wilson method, molecular connection method, and Hansch method is the commonly used 2D-QSAR models. The traditional 2D-QSAR uses structured data that can only reflect the properties of the whole molecule. Therefore, the improvement of structural data and the optimization of statistical methods are of great significance. By improving the structural parameters and introducing new statistical methods, the predictive ability of 2D-QSAR models can be improved. The new statistical methods include genetic algorithms, artificial neural networks, partial least square regression, etc.

3D-QSAR

QSAR is a method of using computational techniques to establish the relationship between the chemical structure of small molecules and their biological activity, which can be used to predict the activity of new chemicals. Using powerful chemometric techniques like PLS, G/PLS, and ANN, the 3D-QSAR extends the traditional 2D-QSAR approaches by making use of the three-dimensional properties of ligands to predict their biological activities. This method can effectively guide the selection of promising compounds and predict the activity of these compounds for structural optimization. The 3D-QSAR model enables researchers to effectively perform force field calculations requiring 3D structures of a training set and reduce the data space by feature extraction and a following machine learning method. The typical workflow of 3D-QSAR modeling includes biological data analysis, 3D structures optimization of biomolecules, determination of bioactive conformation of biomolecules, calculation of molecular interaction energy field, generation and validation of 3D-QSAR model, etc. The 3D-QSAR analysis is a valuable medicinal chemistry tool that can help researchers identify determinants of small molecule bioactivity, optimize existing leads for improved activity, and predict the bioactivity of untested compounds. Our 3D-QSAR models are reliable and validated both internally and externally using rigorous cross-validation techniques.

Fig.2 Structure-activity relationship information obtained from 3D-QSAR study. (Xu, et al., 2020)

Features

• Automatic 3D-QSAR workflow and super high-performance computer
• Fast screening speed and good versatility
• Powerful statistical techniques and comprehensive consideration
• Low-cost and time-saving

References

1. Nantasenamat, C. Best practices for constructing reproducible QSAR models. Ecotoxicological QSARs. Humana, New York, NY, 2020: 55-75.
2. Xu, Y.; et al. 3D-QSAR, molecular docking, and molecular dynamics simulation study of thieno [3, 2-b] pyrrole-5-carboxamide derivatives as LSD1 inhibitors. RSC advances. 2020, 10(12): 6927-6943.