Crowdsourcing anti-malarial drug discovery
Anthony Chubb, Kevin O’Brien, Catherine Mooney, Denis Shields, The FM@H Team*
University College Dublin, UCD-CASL, Dublin 4, Ireland
We propose to perform virtual screening studies against all possible targets in Plasmodium falciparum. Target protein structures include both crystallised
proteins and homology models. Molecule libraries include the freely available ZINC compound library, FDA approved drugs and cyclical constrained
peptides and peptidomimetics. The computational power needed for this endeavour is colossal. We are therefore building a BOINC server that will
harness the world’s unused PC power in screen-saver mode from kind donors. As the computation and analysis is all volunteer based, using open-
source or donated software, we will make all results available to the public. Our aim is that other groups purchase and test the hit molecules, which
should provide novel therapeutic drugs. If you would like to get involved, please go to our website http://bioinfo-casl.ucd.ie/fmah/.
Data Flow Website
The P. fal proteome was searched against
the human, PDB and DrugBank datasets us-
ing the Genomes2Drugs server (http://www.
bioinformatics.rcsi.ie/g2d/) to iden-
tify proteins that had good homology to known
crystal structures, but poor homology to hu-
man proteins. The figure shows the relative
BLASTp scores against the PDB and human
datasets, with the highlighted region indicat-
ing proteins that have not yet been crystallised,
but are likely to be unique to the Plasmod-
ium parasite and are good candidates for ho-
mology modelling. Numerous are putative
or hypothetical proteins that may be of inter-
est but remain unannotated. These will be
cross-referenced against PlasmoDB (http://
plasmodb.org/) and TDRtargets (http://
tdrtargets.org/, 115 PDB structures, 4735
models) to identify target structures to add to
the virtual screening pipeline.
eHiTS docking results for 150k ZINC drug-
like compounds (diversity Tanimoto score 80%)
docked into the crystal structure of the P. fal
FK506 binding protein (2VN1). Insert high-
lights first 100 results.
1 -11.4 ZINC19324669 6 -10.58 ZINC20581662
2 -10.79 ZINC19340398 7 -10.28 ZINC20841257
3 -10.67 ZINC19340382 8 -10.19 ZINC22078620
4 -10.64 ZINC20522231 9 -10.19 ZINC20309805
5 -10.62 ZINC19322221 10 -10.15 ZINC20167871
The first 10 docking results from eHiTS using
a PS3 GRID (TCD). Results are listed by rank,
docking score and ZINC code. Each file con-
tained 5000 molecules, and took 30 hours to
compute on one machine. Total CPU time was
900 hours = 38 days for one machine, but only
2.4 days on 16 GRID machines.
The work was funded through an IRCSET grant to Prof. Denis Shields. SimBioSys for the use of their eHiTS software.
* listed on our website.
The crystallographic position of the native lig-
and FK506 in P. fal FKBP35 is shown with green
carbon atoms. The top hit (ZINC19324669) is
shown with purple carbons.