Dear Ramit Singla,

From my point of view it is impossible to cut the mathematics out of QSAR. However, I would say that there can be two major approaches: statistical and mechanistic.

In mechanistically-based QSAR, you use theoretical knowledge of the problem to select the relevant molecular descriptors (accounting for molecular properties, structural features and other aspects). So you choose a priori what descriptors (your independent variables) to use in the mathematical models.

In statistically-based QSAR, you calculate several (sometimes thousands) of molecular descriptors and then you identify the most statistically relevant for the problem by using variable selection techniques. In this case, it is fundamental that you check that the selected descriptors are relevant not only from a statistical viewpoint, but also from a chemical/biological viewpoint.

For the statistically-based QSAR, you can find some tutorials as well as as references on www.moleculardescriptors.eu, the focus being molecular descriptors (as the name suggests). 

I hope this can help.

Matteo

Dear Ramit Singla,

From my point of view it is impossible to cut the mathematics out of QSAR. However, I would say that there can be two major approaches: statistical and mechanistic.

In mechanistically-based QSAR, you use theoretical knowledge of the problem to select the relevant molecular descriptors (accounting for molecular properties, structural features and other aspects). So you choose a priori what descriptors (your independent variables) to use in the mathematical models.

In statistically-based QSAR, you calculate several (sometimes thousands) of molecular descriptors and then you identify the most statistically relevant for the problem by using variable selection techniques. In this case, it is fundamental that you check that the selected descriptors are relevant not only from a statistical viewpoint, but also from a chemical/biological viewpoint.

For the statistically-based QSAR, you can find some tutorials as well as as references on www.moleculardescriptors.eu, the focus being molecular descriptors (as the name suggests). 

I hope this can help.

Matteo

For the purpose of application, I suggest ChemBench, which was developed by University of North Carolina at Chapel Hill. It is an easy-to-use tool for QSAR modeling. Hope it answers your questions.  

Dear Ramit,

I agree with Matteo that it is quite difficult to totally remove the mathematical part out of learning QSAR/QSPR but we can try to give you a rough idea on what QSAR/QSPR is all about. Here's my attempt. In a nutshell, QSAR seeks to correlate a set of descriptors with a biological or chemical endpoint of interest (i.e. IC50 or Ki for biological endpoint; boiling point, melting point, retention factor, etc. for chemical endpoint). Descriptors or "molecular descriptors" are quantitative or qualitative description of a molecule of interest such as the number of hydrogen bond donors, molecular weight,

In a nutshell, QSAR seeks to correlate a set of descriptors with a biological or chemical endpoint of interest (i.e. IC50 or Ki for biological endpoint; boiling point, melting point, retention factor, etc. for chemical endpoint). Typically, you would begin with assembling or compiling your data set of interest. You can either start with your existing experimental data on the biological activity of a set of compounds that you have synthesized and tested or either retrieving such data from public databases such as BindingDB or ChEMBL. You would then need to either draw or import the chemical structures followed by computing descriptors. Descriptors or "molecular descriptors" are quantitative or qualitative description of a molecule of interest such as the number of hydrogen bond donors, molecular weight, energy of the highest occupied molecular orbital ,etc.  You can calculate hundreds to thousands of descriptors via several free, open source software (i.e. PADEL, CDK, etc.) or commercial software (e.g. DRAGON, CODESSA, etc.). As Matteo pointed out, you can either intuitively select descriptors or have feature selectors pick descriptors to use for predicting the activity. Prediction of activity can be performed using traditional statistical method (e.g. Multiple Linear Regression) or machine learning methods (e.g. Artificial Neural Network, Support Vector Machine, etc.). For this phase of QSAR model construction you can use software such as Weka, R, Python, or ChemBench as Jie Xia pointed out. After you have made the prediction of the activity, you will need to evaluate the quality of the prediction and this is done using several statistical parameters such as the correlation coefficient (R) or R-squared, root mean squared error (RMSE), etc. for regression models where the bioactivity to be predicted is quantitative while Accuracy, Sensitivity, Specificity or Matthews' Correlation Coefficient (MCC) for classifcation models where the bioactivity to be predicted is a categorical label such as Active or Inactive.

For further details, please have a look at the articles listed below.

Hope this helps.

Best regards,

Chanin

Chapter AutoWeka: toward an automated data mining software for QSAR ...

Article A Practical Overview of Quantitative Structure-Activity Relationship

Hi

All of the comment are true, but i offer you if you want to understand the consept of QSAR/QSPR and how to make model, you must read article by prof Alexander Tropsha

.

best regards.

mohammad

https://pharmacy.unc.edu/Directory/tropsha

https://scholar.google.com/citations?user=inPIy48AAAAJ&hl=en