Search algorithms, limitations

To overcome the limitations of the database search methods, conformational search methods were developed [95,96,109]. There are many such methods, exploiting different protein representations, objective function tenns, and optimization or enumeration algorithms. The search algorithms include the minimum perturbation method [97], molecular dynamics simulations [92,110,111], genetic algorithms [112], Monte Carlo and simulated annealing [113,114], multiple copy simultaneous search [115-117], self-consistent field optimization [118], and an enumeration based on the graph theory [119]. 

In summary, pharmacophore model validation is the building of a body of evidence by first relying on validation methods adapted to the search algorithm (and its limitations) and second by resorting to approaches that will incorporate external data and therefore account for some inherent imperfections of the dataset. The first section of this chapter will report some of the most used validation methods and which are related to one or both of these aspects. 

Landscape models are much more abstract than the laboratory technique-based models. As extensive as theory about evolution and optimization on fitness landscapes has become, there is still little work on matching a search algorithm to landscape properties. Additionally, much of this work is based on landscape properties that are presently very difficult to measure with any statistical significance for molecular landscapes. For these reasons, and for reasons of limited space, the landscape search results will be explained in much less detail than the laboratory-based techniques. This section is divided into four parts (i) definitions of terms used in fitness landscape studies and caveats about then-misuse (ii) review of models for fitness landscapes (iii) results from studies of search on fitness landscapes and (iv) conclusions from these results. 

Although in principle a genetic algorithm, or other learning algorithm, shoutd f find the true optimum, the search is limited, either by computer limitations in th i case of numerical studies, or by experimental restrictions in the case of laboratory ir experiments. ij]... 

DST methods are particularly competitive for organic compounds, which are more resistant to the traditional approaches and whose structural models can be easily guessed. At present, the complexity of crystal structures solved by direct-space methods is essentially limited by the number of DOFs that can be handled by the global optimization algorithms within a reasonable amount of time. In prospect, improvement of both search algorithms and computer power may overcome this limitation. The major pitfalls for the use of DST are (a) they are time consuming (b) they are dependent on the existence of reliable prior structural information. Partially incorrect models may compromise the success of the procedure independent of the computer time spent (c) they are sensitive to the accuracy of the peak profile parameterization through peak-shape and peak-width functions. ... 

There are also some other ideas about possible ways to make the calculation of the gradient and of the Hessian more effective, but we limit ourselves to expose topics for which there is a working computer code. The field is in evolution but surely progresses towards computational methods with computational costs and range of applicability comparable to those used for molecules in vacuo are within reach. In our opinion the most difficult point is to find search algorithms for critical points able to treat in a more efficient way some small roughness in the PES introduced by the tessellation of the cavity surface. 

---