Linear-Scaling Ab-Initio Computations of Model Protein Systems - the Potassium Channel and Citrate Synthase

Over the last few years significant advances have been made to develop quantum chemical methods for which the computational time scales linearly with the number of atoms in the system. This project attempts to use these new algorithms in the study of two biologically significant systems, namely the potassium ion channel and the enzyme citrate synthase. Ion channels are responsible for the selective transport of ions across cell membranes, and our aim is to investigate the electrostatic potential inside this system. Citrate synthase catalyzes the condensation of oxaloacetate and an acetyl group attached to the acetyl-coenzyme A. This reaction is an essential step in the Kreb cycle for oxidation of fuel molecules. Use of the APAC National Facility is required since although the basic algorithms are linear scaling the threshold for on set of linear scaling is large. That is to reach systems of a size where linear scaling algorithms become applicable still requires substantial computational effort. Moreover computations on systems of this size require the large memory and disk available on the APAC National Facility. Our work, as well as improving the basic understanding of two biologically important systems, will also involve substantial further development and parallelization of the basic linear scaling electronic structure codes. This is expected to be of widespread benefit beyond this immediate project.

Principal Investigator

Alistair Rendell
Computer Science, Faculty of Science
Australian National University

Project

x54

Co-Investigators

Ryan Olson
Dept of Chemistry
Spedding Hall 201

Joseph Antony
Engineering, FEIT
Australian National University

Andrey Bliznyuk
ANU Supercomputer Facility
Australian National University

RFCD Codes

250601, 280301, 280210

Significant Achievements, Anticipated Outcomes and Future Work

For the first time ab-initio computations using a double-zeta basis set were performed on fragments of the potassium ion channel containing up to 1000 atoms. Both the well-established Hartree-Fock and relatively new B3LYP density functional methods were used. The result of these computations showed that neglect of electronic polarization in molecular mechanics, a technique that is widely used for study of large biological molecules, leads to substantial absolute errors in both the electrostatic potential and energy of binding of K+ ion to the channel. What is perhaps more interesting is that our calculations show that contributions to the binding energy is qualitatively wrongly predicted by molecular mechanics. Ab-initio calculations predict a reduction in the contribution to binding from residues located at a far distance from the potassium ion as system grows in size. In molecular mechanics this is not the case

Another extremely important finding arising from the computations is that density functional methods seem unable to constrain the electron density and predict it to be more evenly distributed over the entire biomolecule compared to Hartree-Fock computations. We are continuing these calculations on some model water clusters.

During our investigation of the potassium channel, some questions arose concerning the validity of commonly used procedure for molecular mechanics parameterization. As the result, we began to extend our initial project to look at the quality of the 6-31G* electrostatic potential for some smaller systems. This work is incomplete, and will be one focus of our work in the coming year.

Related work has focused on the development of efficient parallelisation schemes for performing large scale quantum chemistry calculations. In this work a new implementation of the distributed data interface used by the GAMESS quantum chemistry code was made. This showed significantly better performance on clusters of shared memory parallel computers (like the APAC National Facility SC system) than the previous implementation. Future work on this project is likely to involve tailoring the implementation to include the concept of groups of processors.

 

Computational Techniques Used

Most of the computations have been performed using the Gaussian quantum chemistry package, parallelised using OpenMP. Some computations used GAMESS, which has been parallelised using a distributed data interface

 

Publications, Awards and External Funding

External Funding and Awards

This work was partially supported by the APAC Computational Chemistry Expertise program, and an NSF travel grant.

Publications

R.M. Olson, M.W. Schmidt, M.S. Gordon and A.P. Rendell, "Enabling the Efficient Use of SMP Clusters", Proceedings of SC03, 2003
A. Bliznyuk and A.P. Rendell, "Electronic Effects in Biomolecular Simulations: Investigation of the KcsA Potassium Ion Channel", J. Phys. Chem.(submitted)