ORCO: an entropy based adaptive residue-type classifier for partial protein NMR chemical shift observations

Residue typing of a spin system based on partial solid-state nuclear magnetic resonance (ssNMR) chemical-shift data is often ambiguous because multiple amino-acid types can exhibit similar chemical shift profiles. We introduce ORCO, an entropy based adaptive residue-typing method that uses residue-type- and secondary-structure-specific chemical-shift means and standard deviations to calculate a ranked probability distribution over the 20 canonical amino-acid types. Given the chemical shifts of a spin system, ORCO uses the entropy of the resulting probability distribution to dynamically construct a set of plausible residue types, with the set size adapting to their ambiguity. We evaluated ORCO using 101 high-quality protein data containing 14,519 residues. Using only Cα and Cβ chemical shifts, the ORCO prediction sets contained the correct residue type for 88.7% of residues, with a mean set size of 3.31 residue types. Including N and CO chemical shifts reduced the mean set size to 2.75 residue types while maintaining 88.7% coverage. ORCO therefore provides uncertainty-adaptive residue-type predictions, a prerequisite step for automated sequential resonance assignment, particularly useful for large proteins studied by ssNMR.