A Von Mises-Fisher Clustering Modeling for the Inference of Orientational Preferences of Protein Domains with Scattering Data

• Activity Number: 321 - Machine Learning and Variable Selection
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Seyed Morteza Najibi* and Ingemar Andre
• Companies: Lund University and Lund University
• Keywords: Von Mises-Fisher; Mixture Model; Experimental Curve Fitting; Protein; SAXS

Abstract

Consider two beads connected by a thread, such that the position of the second bead in relation to the first one changes over time with magnetic preferences to limited orientations. If we have an experimental method that measures the average of the relative orientation over time (Small-Angle X-Ray Scattering, SAXS), then we are interested in a model that describes the relative position of the second bead as a mixture of the angular probability distribution over the sphere. Beyond the experimental data, we have a precalculated samples of bead models with calculated experimental data. For this purpose, we proposed a directional probabilistic model that provides inference by clustering the precalculated samples. The proposed approach involves the continuous conformational space of the second beads in estimating the weight function and selecting a set of orientational preferences using the parameters of a mixture of von Mises-Fisher distribution. We are motivated to solve this problem by a structural biology application to characterize multidomain proteins by inferring about the flexible linkers (second bead).