115 / 161
Conformational Ensembles from Experimental Data
and Computer Simulations
Poster Abstracts
111
76-POS
Board 36
A Multi-crystal Parameterisation Method for Separating Atomic and Molecular Disorder
in Crystallographic Experiments
Nicholas M. Pearce
, Piet Gros.
University of Utrecht, Utrecht, Utrecht, Netherlands.
Diffraction experiments result in a temporal and spatial average over many molecules in a
crystal. Atomic Displacement Parameters (ADPs) model harmonic deviations from the average
coordinates arising through thermal motion or crystal imperfections. The ADP for a particular
atom therefore comprises contributions from multiple sources, including: crystal-dependent
disorder; collective, molecule-dependent, “rigid body” motions; and any individual motions of
the atom. Large-scale crystal-dependent factors can be considered an artifact of the
crystallographic experiment, but collective and individual motions of atoms within a crystal may
reveal subtle and biologically relevant protein motions.
Translation-libration-screw (TLS) models are well-established as a method for describing
collective motions of groups of atoms in a crystal. In general, however, the separation of the
overall observed motion into the different contributions (crystal, rigid body, atomic) is
ambiguous, since crystal-dependent factors cannot be uniquely separated from crystal-
independent factors. Furthermore, overfitting is ever-present, and the complexity of a ADP
model is dictated by the resolution of the crystallographic data.
To overcome the intrinsic obstacles of parameterising disorder in a single crystallographic
dataset, we present a multi-dataset ADP-parameterisation approach for modelling atomic
disorder: by characterising the ADPs across a series of datasets simultaneously, using a series of
TLS models, we separate crystal-dependent and crystal-independent parameters. This results in a
hierarchical model of motion, allowing e.g. atomic motions of a sidechain to be decoupled from
the large-scale motions of the whole molecule. This approach is validated by both a reduction in
the R-free/R-work gap across the set of datasets and a decrease in R-free: the multi-dataset
parameterisation thus not only limits overfitting, but increases overall model quality.