Starting with QSAR: What You Need

Steps in QSAR Model Generation:

1. Preparation of input data (structures, known biological activities)
2. 3D Geometry optimization (conformation generation, alignment)
3. Calculation of descriptors
4. Statistical Analysis (Feature selection, regression)
5. QSAR model building
6. Interpretation, validation and prediction

Structure optimization

1. 100,000 atoms: Molecular mechanics - Use of empirically derived potential function
2. 1000 atoms: Semi-empirical quantum mechanics - Use of approximate Schrodinger equation
3. 100 atoms: Ab initio quantum mechanics - Solve exact Schrodinger equation

Descriptor Calculation and Statistic Software for QSAR:

1. ADAPT
2. TSAR
3. SciQSAR
4. Cerius2

Statistical Packages

1. SAS
2. SPSS
3. Minitab
4. STATISTICA
5. SYSTAT
6. StatView
7. WinNN (Neural Networks)

Conformational Search:

1. Grid Search
2. Random Search
3. Boltzman Search
4. Systematic Search

Alignment of Molecules

1. RMS atoms alignment - pairwise model alignment based on superimposition
2. Moments alignment - using electrostatic moments or principle moments of inertia
3. Field alignment - maximizing the overlaps between steric and electrostatic fields calculated using probe potential

Selection of Descriptors

1. QSAR model should be reduced to a set of descriptors which is as information rich but as small as possible
2. Rule of thumb: 5-6 structural points should fall per structural descriptors
3. Objective selection
   1. Correlations
   2. Pairwise selections
   3. Identical tests
   4. Vector space descriptor analysis
4. Subjective selection
   1. Descriptor selection considering biological activity
   2. Genetic algorithm based feature selection

Statistics in QSAR

1. Multiple linear regression:

• Least square error minimization
• N, ANOVA, R, F test, p value
• Examine multi-collinearity: tolerance=1-R2, VIF=1/(1-R2)
• Yields linear models with coefficients that can be interpreted for relative importance

1. Step-wise multiple linear regression

• Forward, backward, stepwise

1. Principle components regression analysis
2. Partial least square analysis
3. Artificial Neural Network method
4. Genetic function approximation
5. Principle component analysis
6. Factor analysis
7. Discriminant analysis
8. Cluster analysis