Date |
Apr 14, 2017 |
Speaker |
Professor Nicolas Wicker, Lille 1 University, France
|
Title |
Multidimensional fitting and projection under pairwise distance control
|
Abstract |
Data analysis is a pretty old field but it is always
possible to have new insights in it.
Here we present two methods that can be useful in many fields but
which can be particularly useful in biology which abounds in complex
and varied data. Firstly, we show an adaptation of multidimensional
fitting to detect protein movements. The core idea is to start from
two
protein conformations, one described by a distance matrix and the
other one by a coordinate matrix. Coordinates are then modified to
obtain approximately the same distances while penalizing movements to
avoid trivial solutions. Secondly, in an analogous way we modify
coordinates and add a radius parameter to each point so that we have
an indication of the approximation error made for each of them, this
is particularly useful for dimension reduction and visualization. This
work fills a gap in the range of methods close to principal component
analysis.
|
|