Toward Fully Automated Drug Discovery May 28, 2015!
Lawrence Livermore National Laboratory is operated by Lawrence Livermore National Security, LLC, ...
Toward Fully Automated Drug Discovery May 28, 2015!
Lawrence Livermore National Laboratory is operated by Lawrence Livermore National Security, LLC, for the U.S. Department of Energy, National Nuclear Security Administration under Contract DE-AC52-07NA27344. LLNL-PRES-668620!
Felice Lightstone!
We need better in-silico prediction tools to reduce the cost and failure rate of drug discovery § 7-15 years and >1$B USD per drug § >90% failure at late stage § Post-market failure due to adverse drug effect (e.g. Vioxx, Lipobay).
2
Computational chemistry tools can focus experimental efforts and resources towards more promising compounds Create 100s of Millions of Drug Candidates
protein
compound
✗! ✓! ✗! 1 Drug 3
High performance computing can speed up drug discovery by increasing sampling of ligand-target complexes § 1 ligand docking into 1 target takes ~1 min. Rescoring 1 pose takes >10 min. § Drug-like compounds ~1060 possibilities. § Druggable targets ~10% human genome/ off-target prediction. Number of PDB Entries
Yearly Growth of PDB Structures 120000 100000 80000
Yearly Total
60000 40000 20000 0
Year (update to March, 2015) 4
We have created an automated system to predict binding in proteins Compound Library
Protein Preparation (PrepP)
Binding Site Evaluation (CatsId)
Docking Studies (VinaLC)
Free Energy Simulations (SimuP)
D. A. Kirshner, J. P. Nilmeier, and F. C. Lightstone (2013) Nucleic Acids Research, 41, W256-W265 J. P. Nilmeier, D. A. Kirshner, S. E. Wong, and F. C. Lightstone (2013) PLoS ONE, 8(5), e62535.
5
A binding site from the Atlas contains residue ids and distance matrix information
File: pdba.dist Pdb: pdba.pdb EC# 1.2.3.4567
Target Protein
The target protein is compared to each Atlas entry Binding Site from Atlas 6
File: pdba.dist Pdb: pdba.pdb EC# 1.2.3.4567
Target Protein
Binding Site from Atlas
All possible combinations of residue binding sites are constructed and compared to the atlas protein 7
File: pdba.dist Pdb: pdba.pdb EC# 1.2.3.4567
Target Protein
Binding Site from Atlas
All possible combinations of residue binding sites are constructed and compared to the atlas protein 8
The best fit for the atlas entry is identified and stored.
File: pdba.dist Pdb: pdba.pdb EC# 1.2.3.4567
Target Protein
Binding Site from Atlas
All atlas entries are ranked according to goodness of fit, and the best scoring entries are chosen as EC# predictions 9
Our automated system of in-house docking & free energy rescoring protocol saves time and increases search space Target | Compound | Complex Slave
Slave Master
…
…
… Output
HPC:! PC:!
~10 min @ 1 CPU! ~10 min!
~1 hr @ 100 CPU! ~4 days !
~1 hr @ 700 CPU! ~1 month!
…
Slave
~5 hr @ 15K CPU! ~10 years!
10
The VinaLC docking program has great parallel performance
Scale to a large set of CPUs (> 15K cores) with low overhead (