Predicting Connectomes Using Noisy and Incomplete Data

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Tract tracing is a technique to show pathways of connections in the brain. The technique is invasive and therefore very costly and time consuming. Furthermore, not all results obtained with the technique are perfect. Therefore, a lot of tract tracing data is incomplete and/or noisy. By modelling the data with the latent space algorithm, we can predict unknown data and reduce the errors in the known data. The macaque tract tracing data of Markov et al. (2012) and Felleman & Van Essen (1991) are both incomplete. Furthermore, their claims about the density of the macaque’s connectome are inconsistent with each other. By fitting a latent space model on the data, the unknown data can be estimated and some of the inconsistencies in the claims can be interpreted. The mouse tract tracing data of Zingg et al. (2014) consist of two different observations of the same connectome. The data, however, is the same for only 79% of the connections. By designing a new type of latent space algorithm, able to fit ordinal data, the two observations are modelled. By fitting the latent space to both datasets simultaneously, it is possible to merge the datasets and form a ground truth connectome. The algorithm used has yet to be perfected, since the found connectome shows some inconsistencies with the expected results. However, the likelihood and accuracy results of the fitted latent space model indicate the usefulness of this kind of models for the future.
Faculteit der Sociale Wetenschappen