Contact Lens: Evaluating Protein Structure by Contacts

1
Contact Lens Evaluating Protein Structure by
Contacts
Tim Dreszer -University of California at Santa
Cruz Rapid evaluation of the similarity of two
structures is an essential tool in protein 3D
structure prediction. Perhaps the most widely
used tool is RMSD, yet it suffers several major
shortcomings. Another is GDT (as used in CASP6),
however it is computationally expensive and still
comes up short. In this project a new tool is
developed based upon residue-residue contacts.
While the basic contact score has strengths, it
is improved upon by rewarding distant contacts
and near identical contact distances, as well as
smoothing the boundary at the contact threshold
and normalizing the score. This contact measure
dubbed contact lens is used to evaluate CASP6
structure predictions. Contact lens proves better
than both RMSD and GDT at resolving some
structure similarities.
RMSD vs. Contact Lens Root Mean Square Distance
is a measure of the distance between the same
residue in two different structures that have
been superimposed. While it is easy to
calculate, it fails to recognize overly compact
structures and allows small discrepancies to
overwhelm the score. A traditional contact
measure is a count of intra-chain Ca pairs that
are within a certain threshold distance of each
other, as compared with the expected contacts in
the native structure.
Conclusions Contact lens shows strong localized
differences from both RMSD and GDT, yet trends
with each of them across the scoring range. The
advantages of Contact Lens over GDT appear most
obvious, as computation is significantly reduced
and intra-model consistencies are rewarded. In
almost all cases where GDT and Contact Lens
disagreed, RMSD agreed with Contact Lens. Like
RMSD, the current form of contact lens may be
overly sensitive to global folding, at the
expense of secondary structure. Fortunately,
this tool is easily recalibrated for near or
far-sightedness. Acknowledgements Kevin
Karplus, fellow students of BME220 and the many
contributors to protein structure prediction
methods at UCSC.