B. Firdaus Begam ¹, J. Satheesh Kumar ²

Affiliations
1 Department of Computer Applications, Bharathiar University, Coimbatore – 641046, Tamil Nadu, IN
2 Department of Computer Applications, Bharathiar University, Coimbatore – 641046, Tamil Nadu

Abstract

Background/Objectives: Quantitative Structure-Activity Relationship (QSAR) / Quantitative Structure - Property Relationship (QSPR) model is based on changes in molecular structure that would reflect changes in observed biological activity or physico-chemical property. Methods/Statistical analysis: QSAR/QSPR involves chemistry, biology and statistics fields for analysis. It has been widely accepted model for predicting association between molecular structure and its activity. Over the years many algorithms have been proposed and applied in QSAR/QSPR studies. Framework of model involves molecular structure (graph) representation, calculation of molecular descriptors (graph invariants) and multiple linear regression method is applied for analysis. Model has been validated through statistical parameters (R and R2). Findings: Methods involved in model development were reviewed for QSAR/QSPR studies. Multiple Linear Regression is one of the best methods for developing QSAR/QSPR model. This work focuses on developing QSPR model for predicting boiling point of alkyl benzene molecules using Multiple Linear Regression method. Wiener index, Harary Index, Hyper Wiener Index, Hyper Harary Index, and Randic index are calculated for analysis. The model has been validated by calculating R and R2 value. Various models were developed based on different combinations of descriptors to analysis which contribute best in predicting boiling point. Best fit model has been identified by developing model with different combinations of descriptors and rank them based on highest R and R2 value. Model with highest value has been taken for prediction of boiling point as best fit model as n (number of molecules) =14, R= 0.9934 and R2=0.9968. Applications/Improvements: Review on methods involved in prediction analysis has enlightened that model with reduced molecular descriptor subset and outlier detection method shows better performance by improving quality of the dataset Main application of QSAR/QSPR analysis is in drug discovery process. As it has reduced the time taken for lead identification and optimization in drug discovery process.

Keywords

Descriptor, Descriptor Selection, Mathematical Model, Multiple Linear Regression, QSAR, QSPR

Full Text