STATISTICAL METHOD OF QSAR
- 1. STATISTICAL METHODS OF QSAR Rani T. Bhagat M . Pharmacy, (Pharmaceutical Chemistry) 1
- 3. 3 Statistical method are mathematical formula, model and technique that are used in statistical analysis of research data. QSAR model represent the mathematical equation correlating the response of chemical (activity or property ) with their structural and physicochemical information in form of numerical quantities i,e descriptor. Regression based approach are employed data of chemical are entirely numerical i, e quantitative or semi-quantitative chemical response are modulated using classification technique Developed QSAR model are also subjected to several validation test to check for reliability of developed correlation method. After it’s development ,QSAR model is usually verified by multiple statistical validation strategies estimation of predictivity and stability. Statistical tools used for data pre treatment feature selection , model development , validation of QSAR . Computer machine learning based method are also useful in developing QSAR model. INTRODUCTION
- 4. METHODS 1) Chemometric tools: Various chemometric tools in QSAR Pre-treatment of data table Features selection Multiple linear regression Partial least square Cluster analysis 2) Quality metrics: Important metrics for determination quality model QSAR Types of validation Validation metrics for regression based QSAR model Validation metrics employ in classification based QSAR Parameter for receiver operating (ROC) characteristic analysis 4
- 5. 1) Chemometric tools Various chemometric tool used in QSAR 1) regression based approach a)Multiple Linear Regression (MLR) b)Partial Least Square (PLS) 2) classification based approach a)Linear Descriminant Analysis (LDA) b)Cluster analysis (CA) Pre-treatment of data table molecular str. Correctly draw Biological activity or other activity have been taken from authentic source Descriptor value have been computed using validate software Response data for QSAR pattern modelling normal distribution pattern Care shoud also taken to avoid duplicate in data set Computation 3D descriptor optimization carried out 5
- 6. Features selection: • Selection of appropriate descriptor for model development from pool of large no. of descriptor is an imp.step in QSAR modelling. • Selection done by variety of ways Stepwise selection – partial F- statistic = ‘F’ for inclusion and ‘F’ for exclusion Multiple Linear Regression: It is used in QSAR due to its simplicity ,trasparency, reproducibility, interpretability. Y= a0 + a1 × X1 + a2 × X2 + a3 × X3 +…………+an× Xn Where, Y-response Dependent variable a0-constant term X1,X2,Xn-descriptorindependent variable a1,a2,a3-regression coefficient 6
- 7. Partial Least Square: • It is better choice over MLR , PLS being generalization of MLR. • It is used for predicting the pharmacokinetic, Pharmacodynamic , Toxicological property from structure derived physicochemical and structural features. • These method developed using the regression analysis. Linear Descriminant Analysis • LDA separate two more classes of object used for classification problem. • LDA show the diff between classes of data predicted membership is calculated by computing a discriminant function (DF) score. • DF value smaller than cutoff value DF= C1× X1 + C2 × X2 +……….+ CM × XM+ 0 Where , DF- Discriminant function C-Discriminant coefficient X- responding score foe variables a- constant m-No. of predictor variables 7
- 8. Cluster Analysis: • Cluster defined through analysis of data. • Cluster analysis maximizes the similarity of cases within each cluster . • And maximizes the desimilarity between groups that initially known. • It is start with each case separate cluster and then combines the cluster sequentially reducing no. of cluster at each step only one cluster is left. DENDOGRAM Cluster 2 Cluster 3 Cluster 3 Cluster 1 8
- 9. 2) Qualitymetrics Important of metrics for determination of quality of QSAR models • Advancement in fast and economical computational resources make it feasible to compute large no. of descriptor using bvarious software. • QSAR model used to check its predictivity for new untested molecule . Types of validation • OECD Principle – Principle 1 Principle 2 Principle 3 Principle 4 Principle 5 • Internal validation • External validation 9
- 10. 10
- 11. Validation Metrics For Regression Based QSAR 1)Metrics for Internal Validation = • Leave –one-out (LOO) Cross Validation • Leave –many-out (LMO) Cross Validation 2)Metrics for External Validation Validation Metrics Employed in Classification Based QSAR Validation Metrics can access the performance of classification – based model in terms of accurate quantitative prediction of dependent variables. Parameters for = 1) Goodness of fit quality determination 2) Model Performance Parameter a)True Positive (TP) b) False Negative (FN) c) False Positive (FP) d)True Negative (TN) 11
- 12. Parameter for Receiver Operating Characteristic (ROC) Analysis 1) ROC Curve TP rate- True Positive Rate on Y-axis FP rate-False Positive Rate on X-axis 2) Metrics for pharmacological Distribution Diagram (PDD) a) Activity Expectancy b) Inactivity Expectancy Activity Expectancy= Ea = % of actives % of inactive + 100 Inactivity Expectancy= Ei = % of inactives % of actives + 100 12
- 13. 13 IMPORTANCE It is used in Computational Chemistry represent molecular structure as numerical model stimulate their behaviour with the help of quantum mechanics . It can Compute energy related properties such as electronic , spectroscopic properties for molecule. It is used for prediction of Constitutional Descriptor , molecular weight , counts of atom,bonds and rings ,topological descriptors, connectivity of molecule. One of most significant and widely used method is using software computed descriptor in QSAR technique.
- 14. 14 Equation generatedestablished in QSAR studies are linear regression equation. A number of equation may be generated or established for one problem case under study. Statistic also help in selecting one suitable best fit equation out of them. This may be done by checking std. deviation or variance and other related statistical parameter for data set used for QSAR studies series of compound. Correlation coefficient computed for data set under study also help in selecting appropriate QSAR equation. Application of Statistics
- 15. 15