journal menu
![[Figure 1]](rr5195fig1mag.jpg)
|
|
Figure 1
Integrand shapes for the acentric and centric distribution for different parameter settings show the variety of function shapes that occur when computing the marginal likelihood. When the experimental error is relatively large with respect to the intensity, high-mass areas of the function span a decent portion of the integration domain for E ≤ 6 (a). When the error on the experimental data is relatively small, the bulk of the integrand mass is concentrated in smaller areas (b, c). In the case of a t-distribution-based noise model, the tails of the distribution are lifted compared with the normal noise model. The variety of these shapes makes the uniform application of a standard quadrature or Laplace approximation inefficient and suboptimal. |
![[Figure 1]](rr5195fig1.jpg)
 | STRUCTURAL BIOLOGY |
ISSN: 2059-7983
access
