Potential of Mean Force of Ion Permeation through alpha7 nAChR Ion Channel September 29th 2009 Dr. Jens Krüger AK Fels University of Paderborn Outline • Neuroreceptors • Ion Permeation • Simulation Performance • Wishlist http://oreilly.com/medical/bipolar/graphics/bpd0102.gif Synaptic Cleft 50 nm nAChR based on Unwin, N. (2005) J. Mol. Biol. 346(4): 967‐989. Transistor Source Gate Drain Receptor (nAChR) Extracellular Neurotransmitter Binding of two agonist molecules (e.g. acetylcholine) to the extracellular part triggers the opening of the channel, wich is over 20 Å away. Intracellular Single Channel nAChR Shao, Z., Yakel, J. (2000) Journal of Physiology 527:3 507‐513. Water density unbound bound Plasticity Changeux, J.‐P., Danchin, A. (1976) Nature 264: 705‐712. Ion Permeation SMD Pulling • virtual spring – moves from point A to B – takes a certain time – has a certain spring constant • object pulled – has a defined starting point – follows the spring – can be faster than the spring • reference group – all movements of the spring are relative to it – good to compensate for small movements in the total simulation system Umbrella Sampling • virtual spring – located along the reaction coordinate – does not move – has a certain spring constant • object constrained – has a defined starting point – follows the spring – can move above or below the spring • reference group – all movements of the spring are relative to it – good to compensate for small movements in the total simulation system Potential of Mean Force SMD pulling Umbrella Sampling Nz PMF ( z ) = ∑ ( ΔFn − γ Δzn / Δtn )Δzn PMF ( z ) = − k BT ln( P ( z ) / P0 ) n =1 Treatment of friction is highly important • 1 simulation (2100ps) yields 1 PMF • 12 repititions for reasonable error estimate • Sensitive to starting configuration and ion placement • 300 simulations (300ps) yields 1 PMF • 3 reptitions for reasonable error estimate • PMF ( zstart ) = PMF ( zend ) = 0 Hydration Shell SMD PMF influx eflux Pulling Speed US PMF Clusters and Middleware • Bisgrid 2 quadcore Intel X5355 (2.66 GHz) per node, Infiniband, Unicore 5.6 (http://www.unicore.eu/) • Arminius 2 singlecore Intel XeonDP (3.2 GHz) per node, Infiniband, CCS (https://www.openccs.eu/core/) Performance Bisgrid 2 quadcore Intel X5355 (2.66 GHz) per node, Infiniband Arminius 2 singlecore Intel XeonDP (3.2 GHz) per node, Infiniband Speedup Bisgrid 2 quadcore Intel X5355 (2.66 GHz) per node, Infiniband Arminius 2 singlecore Intel XeonDP (3.2 GHz) per node, Infiniband Relative Performance Bisgrid 2 quadcore Intel X5355 (2.66 GHz) per node, Infiniband Arminius 2 singlecore Intel XeonDP (3.2 GHz) per node, Infiniband Comparison based on Arminius Steered MD • 2x 12 repititons • 2100 ps length each • 18 h on 32 cores • 750 MB output Umbrella Sampling • 300 repititions • 300 ps length each • 7.8 h on 4 nodes • 150 MB output • 6490 CPU hours are required for 2x 12 runs • 9360 CPU hours are required for 3x 300 runs Wishlist for a Portal • • • • • • • • Combined workflow (command‐line and GUI) Transparent queuing system Multiple input sources Intelligent parameter sweep Handling of large amounts of data (> 1 GB) Flexible data storage Continious progress report Real time application output Acknowledgment Paderborn • Gregor Fels • Brigitta Elsässer • Markus Tusch • Lars Haller • Iris Baum • Oliver Stücker • The Paderborn Center for Parallel Computing (PC2) • Alexander‐von‐Humboldt Foundation • Taiwanese Government, National Science Council and NYMU