Nuclear magnetic resonance (NMR) spectroscopy allows scientists to study protein structure, dynamics, and interactions in solution. A necessary first step required for such applications is determining the resonance assignment, mapping spectral data (e.g. figure above left) to atoms and residues in the primary sequence. Automated resonance assignment algorithms rely on information regarding connectivity (e.g. through-bond atomic interactions) and amino acid type. We are pursuing a graph-based approach (figure above right) that abstracts possible connectivity information as edges between residue vertices, based on possible matches between spectral peaks. The edges in the graph, in combination with sequence information, then drive the assignment process. Unfortunately there is a great deal of ambiguity in edges; the above graph illustrates a correct subgraph (red) obscured by some such noise (black).

We are studying several problems and approaches in automated NMR resonance assignment. We are developing a random graph model capturing the correlation structure in ambiguous edges, and we are using this model as the basis for an efficient algorithm for characterizing, from connectivity information alone, consistent sets of connected fragments. Further, we are studying the combination of connectivity and sequence information and developing a Bayesian statistical model that properly accounts for sources of uncertainty. This model allows quantification of the overall amount of information used, which can vary widely across experimental labs. By measuring the overall plausibility of assignments, we can base our analysis on all assignments (not just the optimal one) that are supported by the data. Finally, we are pursuing an approach based on connectivity information in the unassigned NOESY (through-space) spectrum; for example, the above figure actually shows a portion of the connectivities in a beta sheet, as detected by the Jigsaw program. Our work here generalizes sequential connectivies to higher dimensions (2D and even 3D).

• Noah Benson
• Sheetal Chainraj
• Olga Vitek

• Bruce Craig (statistics, Purdue)
• Bruce Donald (computer science and chemistry, Dartmouth)
• Carol Post (biology and MCMP, Purdue)