Protein structure determination from nuclear magnetic resonance (NMR) spectroscopy using swarm intelligence

Collectively, social insects are capable of performing complex tasks such as foraging for food, cooperative hunting and nest building that are beyond the capabilities of the isolated individuals. Whilst exploring their environment searching for food, ants deposit pheromones on the ground. Once food is found, a laden ant retraces its steps following the pheromone trail back to the nest. The pheromone signals slowly decay, but those ants which, by chance, find short routes from nest to food and back again will spend a greater proportion of their time along any one section of that route, and thus intensify the pheromone trail. These stronger trails draw in other ants through stigmergy - an innate attraction to the pheromone scent - resulting in a self-optimisation of the shortest routes and thus the development of ideal foraging pathways. Such 'swarm intelligence' - the emergence of collective intelligence amongst groups of simple individuals - has been used in mathematical algorithms to solve complex optimisation problems encountered in communication networks and robotics, but never before has it been applied to structural biology. We want to apply such algorithms to the problem of protein structure determination from nuclear magnetic resonance (NMR) spectroscopy data, a task which typically takes a trained spectroscopist weeks or even months to perform. We have constructed a swarm of biomolecular ants which are capable of exploring their conformational space by molecular dynamics simulations, analogous to a colony of real ants exploring their own physical territory. The ants communicate via a pheromone trail: a global list of ideal interatomic distances that are laid down by the ants following simulation of the NMR data, and that are applied as geometric restraints in the molecular dynamics simulations. If any one of the biomolecular ants encounters a structural element that satisfies the experimental data (such as an alpha-helical turn or a beta-hairpin loop), this knowledge is communicated to the other ants in the swarm via the geometric distance restraints thus encouraging them to adopt the same local conformation. Therefore, over time, the ants cooperate in finding the optimal set of inter-atomic distances and therefore determine their own solution structure. Our initial trials were so successful that we patent-protected the swarm-intelligence NMR (SI-NMR) concept. We now want to develop the technique to the point where it is universally applicable to all proteins, and also to the structure determination of protein-protein and protein-inhibitor complexes. It is hoped that this brand new SI-NMR technology would then become accepted as the state-of-the-art method for protein structure determination in both the academic and industrial NMR communities.

Nuclear magnetic resonance (NMR) spectroscopy holds the unique position amongst spectroscopic and diffraction methods as the only technique that can produce atomic resolution structures of proteins in solution. In de novo protein structure determinations by NMR, the key geometric restraints are distance limits derived from inter-proton nuclear Overhauser effects (NOEs). The donor and acceptor protons in an NOE are identified from the chemical shifts of the NOE spectroscopy (NOESY) cross-peak. When each chemical shift relates to only a single proton resonance, the donor and acceptor protons can be assigned unambiguously. However, when either of the chemical shifts can be attributed to two or more proton resonances, the NOE is said to be ambiguous. To resolve any remaining ambiguity, we need to know the protein conformation, but to determine the conformation we need to assign all the NOEs. This dilemma is known as the NOE assignment problem. To date, the most successful strategies for resolving this problem have used 'iterative assignment' schemes and, in this respect, NMR protein structure determination has changed very little since the structures were solved in the 1980s. Instead, we will use 'swarm intelligence' to solve structures in a single-shot, non-iterative manner. These algorithms originate from studies of the collective behaviour of social insects. Collectively, these creatures are capable of performing complex tasks such as foraging for food and nest building that are beyond the capabilities of the isolated insects. Such algorithms have previously been used to solve complex combinatorial optimisation problems encountered in communication networks and robotics, but have never before been applied to structural biology. Our initial trials were so successful that we patent-protected the concept. We now wish to further develop the technology to the point where it is the state-of-the-art method for protein structure determination by NMR.