Hypothetical Active Site Lattice HASL

HAST was described in 1988, the same year as CoMFA. The two approaches are similar in that each calculates an explanatory variable at intersections of a 3D lattice surrounding superimposed compounds.  

HASL differs from CoMFA in the properties calculated at the lattice points. For shape, HASL uses an indicator variable to indicate whether (1) or not (0) a point falls within the van der Waals surface of the compound. This closely resembles MTD, for which overlap or nonoverlap between ligand and hypermolecule is also expressed by a binary code. 

For electrostatic properties, the ligand atoms are classified according to their electron density. Atoms with low, medium, or high electron density are given values of -1, 0, or +1, respectively. These atom types roughly parallel atomic hydrophobicity, such that atoms with low or high electron density are hydrophilic and atoms with medium electron density are hydrophobic. 

Iterative fitting was the statistical tool first applied within the HASL framework. Briefly, as in earlier methods,the biological potency of each compound is apportioned equally to each of its occupied lattice points. The values at each lattice point are then summed over all molecules to provide an initial guess of the contribution of each point to the potency of any molecule. These contributions are then iteratively refined by feedback corrections of the point-centered partial potency values. The iterative cycles, involving one molecule at a time, are repeated until the average observed-calculated error converges to a minimum. 

Because of the large number of variables and their correlations, the final solution is just one of the many possible good solutions. Furthermore, the results are sensitive to the order in which contributions are refined. To circumvent these limitations, Wiese and Coats combined the HASL molecular descriptors with PLS regression. They found PLS to be superior to iterative fitting. This approach moves HASL closer to CoMFA only the descriptors differ. 

---

The methods applied in recent years by various groups to construct QSAR models for ART and analogues include docking calculations to heme [104, 105], CoMFA [106-109] and hologram QSAR [108] as well as the hypothetical active-site lattice (HASL) approach [107], self-organizing maps of the molecular electrostatic potential [110], quantum-similarity measures [111] and topological molecular connectivity descriptors [112]. 

The hypothetical active site lattice (HASL) method (Doweyko 1988) identifies the points of the network associated to the atoms of the molecule of interest. Then it gives a fraction of the value of the biochemical property to these points. This fraction is characteristic of the analyzed molecule. By repeating the procedure for all the molecules in a given set, some points of the network acquire the summing of the assigned values that are different from null values. These points describe a structure as a map of the active site of the receptor macromolecule that interacts with the effector molecules. 

Ligand structures can be represented by molecular fields (electrostatic or steric), which contain enthalpic contributions to binding when implemented by conventional comparative molecular field analysis (CoMFA) (see Comparative Molecular Field Analysis (CoMFA)). Steric volume incorporated in molecular shape analysis (MSA) (Figure la) is another representation commonly used in SAR studies (see Molecular Surface and Volume) Alternatively, in the hypothetical active site lattice (HASL) approach, molecules are represented by a three-dimensional grid of points (lattice) associated with discrete electronic properties. ... 

The merging of data for a series of known inhibitors results in the construction of a HASL (Hypothetical Active Site Lattice) which serves to quantitatively and predictively model enzyme-inhibitor interaction. Details of the HASL methodology are discussed and the approach illustrated using E. coli dihvdrofolate reductase inhibitors. 

HASL (hypothetical active site lattice) is a lattice-based 3D-QSAR technique that distributes partial activities at molecular lattice points (defined within the van der Waals volume of each ligand). The sum of all points belonging to a particular molecule equals the target property of that compound. Its statistical engine is based on multilinear regression. As in CoMFA, a set of alignment rules (i.e., choice of conformers and superimposition criteria) must be provided, but no other information is used. 

D-QSAR = three-dimensional quantitative structure-activity relationship 3D-SAR = three-dimensional structure-activity relationship CoMFA = comparative molecular fields analysis DG = distance geometry HASL = hypothetical active-site lattice MLR = multiple linear regression MSA = molecular shape analysis PLS = partial least-squares. 

---