Only 1.6% of crystal structures in the PDB were solved at resolutions lower than 3.5 Å. Low resolution phasing and model-building are very difficult, despite the fact that alpha helices are identifiable at 9 Å resolution and beta strands at 4.7 Å. Our accumulated experience indicates that efficient use of low resolution data requires efficient propagation of phase information through every step of the structure solution process, so the phase can become an additional observable in the process of model-building. In addition, when information is sparse, as it is in the case of low resolution data, the detrimental impact of bias introduced at intermediate steps needs to be considered in a manner different from the way methods used for higher-resolution data handle it.
We will show how to decide whether low resolution model-building and phasing are feasible in a particular case and discuss our experience with projects ranging from 3.5 to 4.2 Å, for which atomic models have been built despite data scarcity.