Cresset’s field based fragment replacement software ‘Spark’ was designed from the outset to tie its results to synthetic chemistry. This led to options to limit the nature of new bonds that are formed and to sorting our fragment libraries on frequency of occurrence. These options, combined with the superior electrostatics of Cresset’s XED force field for scoring the fragment replacement, gives novel, synthetically accessible molecules in many projects.
In the latest release of Spark we have expanded the synthetic accessibility concept yet further by the introduction of fragments from specific reagent pools. This enables rapid searching of all the possible R-groups that we could include at a specific position for those that best fit our multi-parameter optimization problem. If you are using the database generator tool then the search space becomes tightly linked to your available chemistry – you create a database for the specific reagents that you have in stock, enabling a search of the chemistry space that you can access today.
Comparing reagent and fragment databases in a case study
In this case study we used Spark’s predecessor (FieldStere) in a virtual thought experiment exploring the possibility of using the tool for growing a fragment bound to P38 kinase. We used two inhibitors that had been co-crystallized and deposited into the PDB as 3K3I and 3ROC as our starting points (Figure 1) and tried to grow the smaller inhibitor to be ‘like’ the larger, DFG-in inhibitor.
Fig. 1 (a) P38 PDB: 3K3I DFG-out inhibitor (green) and (b) P38 PDB: 3ROC DFG-in, hinge-flipped inhibitor (yellow).
Initially the experiment was run starting from the 3K3I derived inhibitor as the fixed component and a dummy atom provided in the ortho position of the tetrahydro isoquinoline (TIC) moiety as the site for fragment replacement. The output field similarity score was weighted 80:20 between these two references derived from the two inhibitors – 3ROC : 3K3I. A protein excluded volume was not used.
In autumn 2011 we were pleasantly surprised and excited to see that the results of this exercise were interestingly very close to a structure deposited by Pfizer (Figure 2).
Fig. 2 left P38 PDB: 2YIS, DFG-out inhibitor (purple) and middle and right two Spark output examples.
Using Reagent Databases to grow fragments
We decided to reproduce this experiment again but this time using the protein from 2YIS as an excluded volume and using the new reagent pool databases in Spark.
Assuming that the same chemistry that was used to construct the Pfizer compound can be applied to the TIC example the searching a database of commercially available thiols should provide readily synthesizable results.
Left Pfizer ligand in PDB 2YIS, right: Spark output example 3 from the thiol reagent pool in the same protein.
The third result in the list is shown in Figure 3. This new design shows a beautiful fit into the P38 protein in keeping with the initial thought experiment. Two further examples are shown below (Figure 4) again from the commercial pool which would also be applicable using the same chemical transformation.
Interesting output was also obtained using different reagent pools e.g. amines and alcohols which would involve the use of different chemical transformations but possibly via the same or a different starting chemical TIC intermediate.
Fig. 4 Left Spark output example 8, right: Spark output example 84 from the thiol reagent pool, both in the same protein.
The new Spark reagent databases enable rapid profiling of the accessible chemistry space around a specific core. Using the Spark database generator enables generation of reagent databases that are linked tightly to your immediately available moves.
To find out more register free for our webinar on April 29th ‘Using reagent databases to find the next move – accelerating lead optimization with bioisosteres‘.