1	Methods for Unknown Receptors Free-Wilson Approach MTD and MSA Methods CoMFA Multiple Binding Modes
2	Model-based QSAR fast formation of the reversible 1:1 complex the effect is immediate consequence of the receptor modification The effect is proportional to the fraction of the receptors modified
3	Estimating Ligand-Receptor Interactions
4	Receptor-Site Models
5	Free-Wilson Method I
6	Free-Wilson Method II common or similar skeleton activity of a molecule is the sum of contributions s of varying substituents a substituent contributes to biological activity independently of its position
7	Free-Wilson Method III
8	Free-Wilson Method IV
9
10	Early Approaches for Non-homologues Limitation to a homologous series was restrictive. It only allowed optimization within the series. A broader perspective was required. The first approaches that broke the limitation of the common skeleton --- MTD (Minimum Topological Difference) --- MSA (Molecular Shape Analysis) relied on a superposition of active compounds.
11	Minimum Topological Difference I
12	Minimum Topological Difference II z = 1 for atoms that bind to the binding site z = -1 for atoms that interfere with the binding site z = 0 for other atoms, e.g. inside the binding cavity but not binding
13	Minimum Topological Difference III The initial assignment is a hypothesis that is tested using the expression The sum includes all f-values for the given molecules. The assignment is systematically changed until a good correlation is achieved.
14	Minimum Topological Difference IV considers superposition although at a very limited basis introduction of combinatorial methods
15	Molecular Shape Analysis superimpose molecules calculate molecular volumes relate the differences in molecular volume to activity
16	MTD and MSA Evaluation both methods consider molecular structure to some extent both methods reduce the 3D-diversity into a single number that they correlate with biological activity MSA uses more realistic superpositions MTD introduces a combinatorial procedure
17
18	Modern Methods I
19	Modern Methods II
20	3D-QSAR: Distance Geometry I subjectively design a collection of binding points create distance matrices for individual compounds (encode conformational flexibility) --- upper triangular matrix contains maximum interatomic distances --- lower triangular matrix contains minimum interatomic distances
21	3D-QSAR: Distance Geometry II compare distances between binding points and atoms using quadratic programming --- each molecule can bind in several binding modes --- the procedure selects the best binding mode for each compound
22	CoMFA: History Richard Cramer III, 1987 started as DYLOMMS (Dynamic Lattice Oriented Molecular Modeling System) combined two existing methods --- GRID (Goodford, 1985) --- PLS analysis patented in 1991 and 1994 commercially available in Sybyl suite (Tripos)
23	CoMFA: Principle I relates differences in biological activity to differences in the shapes of the non-covalent fields surrounding the tested molecules provides a map of the binding site fields steric (Lennard-Jones) electrostatic (Coulomb) hydrophobic - HINT
24
25
26	CoMFA: Statistical Procedure the system is under-determined Partial Least Squares (PLS) finds combinations of the structural descriptors - latent variables (t) that --- are well predicted by the structure descriptors (X) --- predict the biological activity data (BA) well t_i = a₀ + a₁S001 + a₂S001 + …….+ a_kE998 BA = b₀ + b₁t₁ + b₂t₂ + …….+ b_nt_n Y = b₀ + b₁S001 + b₂S002 + …….+ b_kE998
27	CoMFA: Cross-Validation I PLS is working with more variables than data points without precautions, it can easily ‘overfit’ the data cross-validation is used to --- find an optimal number of components (latent variables) to prevent overfitting --- check the predictive ability of the model
28	CoMFA: Cross-Validation II omit one or more rows of input data re-derive model for remaining input data predict the biological activity values of the omitted rows compare the calculated activity with the experimental activity calculate the statistical indices --- predictive least squares (PRESS) --- cross-validated correlation coefficient
29
30
31
32
33
34	4D-QSAR developed by A.J. Hopfinger, University of Illinois, Chicago the 4^th dimension – conformational flexibility generates a set of 3D-models that are valid for individual conformers (binding modes)
35	Quasi-atomistic Pseudo-receptor Models Author Angelo Vedani, Biographics Laboratory 3R, Zurich, Switzerland software Quasar population of 500 or so pseudo-atom models 4^th dimension – binding modes 5^th dimension - ionization
36	Multiple Binding Modes: Interaction Scheme
37	Multi-Species Multi-Mode Binding
38	Fractions of Individual Species
39	Multi-Mode Approach can be applied in any receptor-site modeling method the procedure that is normally used to express the association constant, is now used to express the partial association constant for a single binding mode the results are summed up according to the relations between observed and partial association constants The results include the optimized fractions of individual binding modes, in addition to the standard output – the receptor map and the prediction equation