Specs' targeted sets are selections from our compound collection of diverse and lead-like screening compounds that are targeted at certain activities or therapeutic areas. For the selection of targeted libraries we can use two of our in-house Cheminformatics tools. These are based on two fundamentally different approaches, which are outlined below:
Selection using predictive software for biological
The program calculates a set of descriptors
for each structure in Specs' stock. Next, it calculates a percentage indicative
of the chance that a certain structure will show a specific activity, based on
whether these descriptors are found in the training set. When a search has been
performed, the result is a selection of compounds from our stock that have a
higher predicted chance of being active against selected targets than a random
selection. The threshold for predicted activity against a target can be set
between 0.0 and 1.0; i.e. low predicted activity (<0.3) to high predicted
activity against a given target (>0.7).
List of predefined classes of targets or therapeutic areas:
Examples of some activities in this case associated with the predefined set of Protease Inhibitors:
Please note that any combination of therapeutic areas and/or activities can be selected in order to optimize the hit rate for YOUR specific research area!
Clustering in Specs' chemical space
First, about thirty relevant descriptors are calculated for all compounds to be clustered. The descriptors are designed to capture topological and connectivity information, hydrophobic and hydrophilic effects, polarizability, and electrostatic interactions.
Second, principal components analysis (PCA) is performed in order to reduce dimensionality of the descriptor space and to exclude correlated descriptors. Next, Specs' compounds are clustered together with hit molecules. The hit molecules are either known from literature active agents or proprietary to the customer. Alternatively, the highest scoring molecules from the first approach (predictive software for biological activity, point 2 of this document) can be used as hits for clustering. Ultimately, compounds from clusters of the hits are selected. The clustered compounds can be plotted in the PCA space. ¤
Comparing the two methods
Using the first approach (predictive software for biological activity) molecules structurally similar to known active agents are likely to be predicted as active. This is thanks to the character of the calculated MNA descriptors. Using the second approach (clustering based on property encoding descriptors) selected compounds are less likely to be structurally similar but have resembling electro- and physico-chemical properties with the known active agents.
Some details concerning the Specs' stock
Specs' stock consists of drug-like novel molecules. Majority of compounds in our stock complies with the Lipinskis' and Vebers' rules. Number of physico-chemical parameters like LogP, TPSA, number of hydrogen donors, number of hydrogen acceptors and number of rotatable bonds are in-house calculated for all compounds in our stock. Prior to a targeted selection we can exclude compounds violating any specified criteria.
The targeted libraries are selected from current Specs' stock, which consists of non-virtual, available compounds. Specs' stock is composed of diverse chemical compounds thanks to our global acquisition strategy. To demonstrate the diversity in our stock a fragmentation procedure was run on structures in our 10mg October stock database. 207,978 structures were fragmented. The database is build out of 3992 cyclic scaffolds and 2085 linkers (non-cyclic scaffolds). There are 1733 different cyclic R-groups and 2452 single and multi-atom caps (non-cyclic R-groups).