Protein structural models are useful in a variety of applications. However, not all structures are solved
with a degree of accuracy that makes them suitable for use. And surprisingly, even in “good”
structures, not all regions are accurately solved. Interpreting from unreliable models - or unreliable
regions - yields unreliable results, and models with major structural defects may behave unpredictably
in computational applications. Some form of quality assurance is necessary.
The Richardson Lab provides tools for structure validation, available through the MolProbity
webservice. Recently developed tools expand the range of MolProbity’s validations. CaBLAM
analyzes protein backbone geometry and has proven especially useful in the resolution regime around
3Å, where much recent cryoEM work falls. For higher resolution structures, UnDowser identifies and
categorizes problematic waters. And automated cis-peptide identification has successfully combated a
rise in incorrect cis-peptides caused partly by use of insufficiently filtered modeling libraries.
Accordingly, we are also interested in producing usefully diverse and properly filtered structural
libraries. We are currently releasing the latest edition in our series of high-quality structure databases.
This new database incorporates our latest understanding of residue-level filtering for structure quality.