I am interested in attempting some scaffold hopping type experiments
search2d.py takes a initial model computes its 2d pharmacophore using RDKit. This is then search against a smiles library of compound In practice I have been using the Zinc Clean Leads database but a truncated version is uploaded here for trial (Zinc license prevents distribution of large chunks of Zinc)
Ideally I would parallelise the search but at the moment it is a serial search so is on the slow side. The starting molecule is compared with each member of the database using a Tanimoto similarity on the Pharmacophore fingerprint.
python search2d.py > log.txt
Once we have a list of compounds of similar pharmacophore, brics-scaffold-hop.py takes the compounds fragments them using the BRICS algorithm and then from these generates a new library of compounds. These de novo designed compounds are then assessed in a negative design fashion by comparing both similarity to the target parent molecule and to a known non-specific inhibitor.
python brics-scaffold-hop.py > bricsOut.txt
This script also plots a graph of DeltaSimilarity(Positive design - negative design) vs similarity to positive.
New generated compounds generated show enriched delta similarity compared to a search against the Zinc DB alone.
Things I want to do....
- parallelise search
- use QSAR model to predict affinity of de novo designed compounds
- Multidimensional optimisation
- Genetic algorithm for compound generation rather than BRICS