Talk: Real-Spacing Modeling of Solution State SAXS

2pm Garden Room

Presenter: Dr Robert Rambo Diamond Light Source

Abstract

Small angle X-ray scattering measurements of dilute, homogenous particles
in solution are resolution limited measurements of the thermodynamic
ensemble. Similar to X-ray crystallography and electron microscopy, SAXS
observations made at higher scattering vectors (q) imply a greater detail
in the structural measurement. Here, I present a new ab initio modeling
method that exploits two fundamental properties of Information Theory
(namely, the Shannon Sampling and Noisy-Coding Channel theorems). These
theorems allow for the error-free recovery of the SAXS signal, in the form
of a real-space, cross-validated P(r)-distribution. The P(r)-distribution
contains the structural assessment of the thermodynamic ensemble that is
evenly distributed over discrete points determined by the Sampling
theorem. As such, an adaptive simulated-annealing, bead-density modeling
algorithm is proposed that targets the P(r)-distribution using the
Kullback-Liebler divergence, an Information Theory difference metric.
Compactness is achieved by treating the search space (i.e., the set of
equally sized beads within a hexagonal close-packed lattice) as a convex
set where the minimization seeks to minimize the feasible set of selected
beads (convex hull).

The algorithm scales with resolution. Using a SAXS dataset of a 25
base-pair, double-stranded DNA, the volumetric model illustrates features
of the major and minor groove as the resolution of the SAXS dataset
increases. Further tests on SAXS of the P4-P6 group I intron RNA domain
reveal the large solvent channels observed in the X-ray crystal structure.
This method shows that modeling can be made more reliable by exploiting
theorems from Information Theory.