OpenWorld San Francisco

During the first week of October, I attended the Oracle Open World conference in San Francisco, where I was invited to give a talk entitled “Collaborative Cloud-Based Virtual Reality for Scientific Research & Education”. In the talk, I outlined the virtual reality framework that we’ve been working to develop over the years, focusing on the cloud aspects of the project – particularly those which enable multiple users to simultaneously inhabit the same real-time virtual simulation environment.  Using modern cloud architectures, it’s now possible to build real-time interactive simulations that harness the power of cloud supercomputing. And the cloud allows anybody to login to the simulation server remotely. Another highlight of the trip was an invitation to a small get-together at Larry Ellison’s private SF residence, where he offered his insight into a wide range of different areas, all informed by his perspective as the founder of one of the planet’s biggest tech companies.  Amongst his most memorable statements was his claim that, were he to start over again, he’d consider starting a molecular science or biotech company. He also said that computational molecular biosciences is one of his hobbies. It’s an area he’s always been interested in, but wasn’t really a viable discipline back in the day. Things have definitely changed.

modern art oxford: future knowledge

On 17th June, Lisa May Thomas and I led a workshop at Modern Art Oxford entitled “Sculpting the Invisible World”. The work was part of the gallery’s ‘Future Knowledge’ program of events, curated by Emma Ridgway, and photographed by Stu Allsop. Using a pioneering multi-person virtual reality software framework, visitors were invited to interact within a virtual landscape as embodied energy fields. Methods from rigorous computational molecular physics and real-time digital rendering allowed digitally embodied participants to sculpt the dynamics of a simulated molecular nano-world, for example deforming buckminsterfullerene molecules, passing them back and forth, threading methane molecules through a carbon nanotube, and tying knots in proteins.

Sculpting the Invisible World follows on from the ‘dances with Avatars’ experiments carried out by Lisa May Thomas, which were designed as a sort of embodied Turing Test. During the Modern Art Oxford workshops, we were specifically interested in two aspects of multi-person interaction in VR: (1) what are the conventions which guided human-to-human interaction in virtual spaces, when we are rendered as digital bodies? (2) how do we begin to understand what ‘feeling’ means in an immersive scientific visualisation environment – particularly in order to understand workshop participants’ claims that different molecular structures “feel” different?

Photosynthetic excitation energy transport


Another paper to report on (open-access link available here). This work examines excitation energy transport in LH2, a supramolecular photosynthetic complex which is found in the cell membranes of purple bacteria. Lots of people have gotten interested in LH2 ever since Graham Fleming’s group published a paper in 2007 reporting on fancy 2d spectroscopy which observed coherent quantum “beating” between initially prepared electronic states. Beating patterns of this sort are certainly of fundamental interest, and the experiments used to observe it were very nice; however, the consensus which seems to be emerging is that the “beating” is in fact not so important for explaining the efficiency at which photosynthetic systems transport electronic energy across their membranes.

So why did I decide to get involved in LH2? Well, a few reasons. First, my colleague Dr. Tom Oliver has had a longstanding interest in this system. And second, I’ve been extremely confused by the LH2 modelling literature. When we think about excited state and non-adiabatic dynamics in small and medium sized molecules, we think about topological features like conical intersections, avoided crossings, near degeneracies, non-adiabatic coupling vectors, etc. For small molecules, we know that fluctuations in vibrational degrees of freedom are often responsible for bringing electronic states into near-degeneracy, and transferring amplitude between states.

Despite the fact that all of these concepts are extremely mature owing to developments over the years within the small molecule excited state dynamics community, they are nowhere to be found when one reads the LH2 literature. Instead, we read about master equation treatments where every single vibration is anonymised, and folded into a linearly coupled “spectral density function”. Conical intersections? Avoided crossings? Near degeneracies? Non-adiabatic coupling vectors? Entirely absent. It’s almost like LH2 is linked to an entirely different discipline with an entirely different vocabulary.

So my aim with the LH2 work was full representation of vibrations! No more nameless vibrations folded into some anonymous spectral density function. Full representation of each and every vibration, capturing the fullness of its unique dynamical anharmonic identity. It’s the kind of sentiment that seems particularly well aligned with the current populist zeitgeist sweeping the globe.

But LH2 is big, and it has has lots of excited states, so a full representation of all of its vibrations in atomistic detail is a non-trivial challenge. To do it, we built a multi-tiered parallel computational framework (using TDDFT parallelized across GPUs within nodes, and MPI to scale across nodes) for calculating atomic cartesian gradients on each and every excited state. We also introduced some approximations for how to treat cartesian gradients of the excited state dipole moments and the transition dipole moments. This builds on some work we published a couple years ago. Once all that was stabilised, we were then able to run surface-hopping simulations to treat the explicit dynamics of both the atoms and also the electronic degrees of freedom. The picture that emerges is shown in the movie. The key thing which this movie highlights (compared to previous simulations) is the fact that the atomic motion is explicitly being accounted for, along with the electronic motion (which is shown as diffuse blue clouds). You need to look closely at the video to see the atomic motion, because your eyes are more attuned to the flashing blue electronic amplitude than to the subtleties of the vibrational motion. But it’s definitely all in there if you look closely!

The dynamical picture of electronic energy transport which emerges from this work is one of excited states which fluctuate rapidly as a result of the underlying vibrational dynamics of the atoms which make up the constituent LH2 chromophores. The excited states are delocalized over multiple chromophores and undergo frequent crossing on a femtosecond timescale, as depicted in Fig 4A of the paper. Every crossing offers an opportunity to transfer amplitude from one excited state to another. The result is a sort of highly connected excited state network: the frequent crossings combine to create scenario where the states are in a sort of constant “communication” with one another, allowing excitation localized in any one state to travel far and fast. The take-home message? It’s all about the vibrations!

dynamics at diamond surfaces


We’ve just published a paper (open-access link here) looking at non-equilibrium reaction dynamics at the surface of diamond. As shown in the video, our simulations enabled us to look at the dynamics of a slab of hydrogen capped diamond. We constructed the diamond slab so as to contain a single “dangling” surface methyl group. Diamond is a notoriously good heat conductor: from a microscopic perspective, this means that it dissipates vibrational energy extremely efficiently. So we were surprised to observe that placing some initial heat into the dangling CH3 group (by “plucking” its bonds) led to a statistically significant number of dissociation events. Not an enormous number of events, but enough to matter.

The conventional knowledge is that such events essentially have a probability of zero – owing to the fact that diamond is so efficient at heat dissipation. So we set out to understand why we observed any dissipation. To do so, we formulated an energy-grained master equation model, an area where I’ve been active for while. To some extent, this approach followed on from work we carried out back in 2009, which was aimed at understanding the competition between reaction and relaxation dynamics in organic solvents. Our 2009 study was the first to apply the energy grained master equation to reactions in condensed phases; and to the best of my knowledge, the diamond work represents the first attempt to apply the energy grained master equation to reactions at surfaces.

The big problem with a master equation representation of diamond arises from the fact that it’s not straightforward to separate the “system” degrees of freedom from the “bath” degrees of freedom. For reactions in liquids, the separation is somewhat more straightforward: it is effectively the solute/solvent distinction. For diamond, we simply tried a sensible definition: let the “system” be defined as those atoms and corresponding vibrations which were less than 3 covalent bonds away from the constituent methyl group; and let the “bath” be defined as everything else. This allowed us to calculate energy resolved rate coefficients [k(E)s] for CH3 dissociation from the “system” component of the diamond surface. The next issue a master equation approach faces is this: how do we represent the energy transfer rate from the diamond “system” to the “bath”? Our approach was to run a single long trajectory of the diamond slab (as shown in the video), and then use linear response theory to analyze the characteristic timescale for energy fluctuations within the diamond “system” to dissipate into the “bath”. Then we parameterised an energy transfer function to fit the energy decay curve. With these two ingredients – k(E)s, and an energy transfer function – we could run master equation simulations of the surface reaction dynamics. The results showed the energy dissipation from the system to the bath is definitely fast (with a timescale of ~100 fs) but that there is indeed a non-trivial probability that “prompt” CH3 dissociation events occur – i.e., prior to dissipation of all the energy.

Similar to the 2009 results, we actually found that the master equation did a pretty decent job compared to full MD simulations of CH3 dissociation! But for a computational cost which is 1/100,000 the cost of running the full set of dissociative MD simulations – big savings! As the first of its kind, this work is preliminary in many important respects, but it definitely offers a viable option for relatively cheap modelling of non-equilibrium reaction dynamics at surfaces. Note that all of the input files required to run the model are being made available in MESMER, our cross-platform, open-source master equation solver.

proteins in VR part II: tying knots & binding ligands

A little bit more progress in our molecular VR research work… Building on the framework which we demoed in Salt Lake City at Supercomputing 2016, we’ve started looking at applications to biomolecular systems with interesting conformational dynamics which are difficult to observe using standard molecular simulation workflows. The two videos that I’ve posted here were made by PhD students Mike O’Connor and Helen Deeks. The videos show Mike & Helen’s view within the real-time Nano Simbox virtual reality environment as they utilize a wireless set of “atomic tweezers” to steer a real-time molecular dynamics simulation (i.e., a real-time GPU accelerated implementation of the AMBER force field).

The first video shows the steps which Helen took to tie a knot in a 10-alanine peptide. Knotting is an interesting application for the VR Simbox, because the manipulations required to tie a knot in a molecular structure are actually pretty complex. For example, if I was going to write some code to tie a molecular knot, it would end up being a rather complicated little piece of software. However, tying knots is the sort of thing that’s actually rather straightforward and intuitive for a human, because we all tie knots all the time (and the sailors and knitters amongst us are even more expert)… There’s a lot of fundamental interest in understanding the kinetic mechanism of knotting, given that 1 – 2% of all known proteins are knotted…

The second video shows the steps which Mike took to interactively dock a single benzylpenicillin drug molecule (initially floating in free solvent) into the active sight of the β-lactamase enzyme. β-Lactamases are amongst the most common molecular tool used by bacteria to break down important classes of β-lactam antibiotics like benzylpenicillin, causing them to lose their antibiotic effect. Understanding the mechanism of β-Lactamases is therefore essential to make progress addressing the growing problem of anti-microbial resistance.

In both of these videos, Mike & Helen were able to generate dynamics pathways which would simply never be observed using conventional simulation methodologies. We’re now working on methods for analysing the user-generated pathways – i.e., enabling us to map conformational states, and also to calculate free energies. The idea is that this will provide insight into conformational kinetics and mechanisms. Stay tuned!