Personal tools
You are here: Home Events ANC Workshop Talk: Botond Cseke and Colin Buchanan, Chair: Matthias Hennig

ANC Workshop Talk: Botond Cseke and Colin Buchanan, Chair: Matthias Hennig

— filed under:

  • ANC Workshop Talk
When Jul 17, 2012
from 11:00 AM to 12:00 PM
Where IF G.03
Add event to calendar vCal

Botonde Cseke

E. Coli meets two sparse matrices

In this talk, I will present some of our attempts to do both inference and structure learning when combining microarray and metabolite data resulting from experiments on E. Coli, ie, inferring transcription factor activity and metabolite effects.  On the technical side:  input-output factored hidden Markov models or Gaussian Markov fields and inference with variational Bayes (with a slight twist).


Colin Buchanan

Whole-brain structural networks from diffusion MRI

The structural connectome may be explored at a macroscopic scale in vivo through diffucsion magnetic resonance imaging (dMRI) and whole-brain tractography.  Under this approach, segmented cortical areas (e.g. Brodmann areas) form the nodes of a network and tractography is used to construct a set of white matter fibre tracts which form the network connections.  Graph-theoretic measures may then be used to characterize patterns of connectivity.  However, the test-retest reliability of the resulting networks is currently understudied.

We estimated the confounding influence of three variables, namely, registration error, segmentation error and inter-session differences in tractography computed from ten healthy volunteers who underwent a dMRI protocol on two occasions.  Our method involves parcellation of the cortex from 3D T1-weighted scans into 84 regions-of-interest per subject (FreeSurfer) and tractography initiated from dMRI voxels using an established probabilistic algorithm (PICo).  We guantified the between and within-subject components of four commonly used graph-theoretic network measures.  In all four cases, the within-subject differences were smaller than between-subject differences with an interscan differnece of less than 6%.