|
|
 |
The MATRIX approach increases substantially the probabilities of obtaining successful drug candidates.
Let us illustrate the main differences between a MATRIX-powered drug development and one using “conventional” technologies as follows:
We are, of course, assuming that that there is a solution to our problem, i.e., that there really is a molecule which has the desired impact on our target, is specific in its action, is not toxic, can be brought to the place in the body where it should act, and does not show any undesired side-effect. In principle, we can look everywhere for such a molecule, though the enormous number of possibilities makes it impossible to test all that we would find. Our search is for the optima in a very complex and highly dimensional landscape of the structure-activity relationship and the structure-toxicity relationship and so on.
It should be noted that the final “fitness-landscape” for the compounds, i.e. the quality criterion of all single criteria together, will typically be a completely different shape than all of the single landscapes.
|
|
Figure 3: Schematic illustration of fitness landscapes for two criteria: criterion 1 in the leftmost subfigure, criterion 2 in the middle subfigure and the combined quality criterion (here simply realised by the sum of the both single criteria) in the rightmost subfigure.
|
In order to find lead structures, conventional approaches such as HTS start with a limited set of molecules. In the filtering process following this primary screening it is assumed that no contradictory requirements are made of the compounds. This is not very realistic, as there are invariably further factors, one typical requirement being for a given specificity. It is purely a matter of chance that there are sufficient numbers of hits within the limited primary screening set to find compounds with the desired attributes and with a sufficient space for the variations needed in later phases.
In mathematical terms, this conventional approach works as follows: a very limited subset is scanned, yielding numerous local optima to be further investigated in detail. With every new criterion that has to be taken into account, the local optima from the former tests are taken as starting points for new optimisations with respect to the new criterion. If the successive shifts away from the original optima are not too significant, the process may come up with local optima of the total fitness. This total fitness is the weighted combination of all criteria. The approach fails if two criteria are contradictory, i.e. if, for almost all parts of the chemical space the maximum for one criterion coincides with a minimum for the other criterion. To find the rare regions in space in which this is not the case is very unlikely since the search is more or less random.
The MATRIX approach acts directly on the landscape of the combined requirements. Many regions of the molecular space are circumvented, if they are not promising, thereby considerably reducing the experimental effort required.
More importantly, the workings of the algorithm to only accept solutions that meet the requirements in all respects leads the search process into regions sometimes far removed from those investigated when only taking a single criterion into account .
Fig. 3 shows a typical example of this: the “middle hump” would be typically overseen by conventional discovery approaches.
|
|
