Sire

Big news - I've heard that my paper that describes my new method for calculating absolute binding free energies has been accepted for publication in the Journal of Chemical Physics. I've blogged about the method before, and how the method was one of my main motivations for developing and designing Sire.

I chose to publish in JCP because they allow authors to post copies of papers on personal websites, so I'll post a copy here once it is available. To encourage adoption of the method, Ros (my colleague who helped run the simulations and write the paper) and I have put together a webpage that describes the paper, and also provides the link to the exact software used in the paper, and all of the input files used in the paper. These files are organised by paper figure, so you can download and run the simulations that correspond to each figure (we hope - and expect - you get the same results ;-) - let us know if you don't!). We've done this as we strongly believe that computational chemistry has to do more to become reproducible. It is not right that students today spend many months just trying to reproduce published results. The problem is that, while the paper may describe the software and methods, it is not a complete description. To be able to reproduce results you need the exact input files, and the exact parameter files. Without these, the chaotic nature of the simulations can lead to very different behaviour. Worse still, different, subtle choices (such as arrangement of waters in solvation shells, choice of titration states of residues, exact values of novel forcefield parameters) can lead to different results, or even unstable simulations.

Since the last time I blogged (admittedly nearly a year ago!) Sire has come on in leaps and bounds. I now have several users, and several pieces of work using Sire have now been completed and there are now several papers at different stages of production/submission. I've got two software grant applications in at the moment that use Sire as a base - one looking at developing the work queue / work packet system used by Sire to distribute computational work over compute resources (this is called "Aquire") into a cloud computing platform. The other is to develop an adaptive hybrid dynamics simulation program (called "Aspire") that will use Sire as a foundation. This will be a GPU/CPU, MC/MD, CG/MM hybrid code, and should allow me to move onto the next big science project idea I've had - the development of free energy simulations that include the cost of large-scale conformational change, e.g. calculating absolute protein-ligand binding free energies that include, directly, the cost of changing the protein conformation from the free to bound form. This is one of the last major challenges to solve to allow for the accurate and efficient prediction of binding affinities, and I am confident that the successful completion of Aspire will be a significant step towards making computational chemistry into a tool that will be indispensable across the whole drug discovery pipeline. Well - I can hope - 10 years ago now, as a new PhD student I remember driving back with my supervisor after visiting our industrial collaborator, and we predicted what the future of computational chemistry could be. My prediction was that, while the protein folding problem would not be solved, the protein-ligand binding problem would. Obviously, 10 years has now passed, and while I was right with the first prediction, I was wrong with the second. Protein-ligand binding affinity prediction is still not reliable and efficient. However, with the water-swap reaction coordinate, the RETI free energy method, the QM/MM multiscale Monte Carlo method, and in the next year, the Aspire hybrid dynamics program, I think that incredible progress has been made, and the final solution is in sight :-).