Katharina Heil and Gavin Gray
Tue 18 Oct 2016, 11:00 - 12:00
IF 4.31/4.33

If you have a question about this talk, please contact: Gareth Beedham (gbeedham)

Katharina Heil

“Gaining insights into Parkinson's Disease through the use of Protein-Protein Interaction Networks”

 Parkinson's Disease is the second most common neurodegenerative disease in the modern world. Even though it is extensively studied, causing factors are still unknown. Nevertheless a growing number of genetic alterations have been linked to the disease. This allowed to identify dysfunctioning pathways such as mitochondrial dysfunction, alpha-synuclein misfolding, failure of protein degradation and others. 

Due to the disease complexity, certain combinations of dysfunctions can be found in different patients, defining the individual disease picture.

Since current technologies do not allow to apply experimental approaches to such a large scale problem, computational approaches come into play. We are using protein-protein interaction information to build biological interaction networks. Clustering algorithms e.g. 

spectral community detection, are applied to divide networks into clusters, representing protein-subgroups. Subgroups contain proteins that are more likely to interact amongst each other than with the rest of the network. Functional properties of clusters can be assigned to subsets through Gene Ontology enrichment studies.

We have applied a similar approach for genes associated to Parkinson's Disease. This allows us to detect protein-groups that are overly enriched in disease genes. Correlating Parkinson's Disease gene properties with the dominating functionality of genes in enriched communities, we can confirm the hypothesized disease-complexity. The analysis of associated proteins that are not (yet) linked to Parkinson's Disease gives further insight and can hint towards future drug targets and markers to be used in diagnosis.

In my talk I would like to introduce our network approach, the used datasets, outcomes of the enrichment studies and draw first conclusions about the newly gained insights to Parkinson's Disease.

 

Gavin Gray

“Resource-efficient Feature Gathering at Test Time”

Feature gathering refers to the process of gathering samples that will be combined to create features to be used in the traditional feature vector to use in a supervised learning task. At test time, we posit a price on getting each of these samples, and define this task to find the most efficient allocation of resources in order to maximise a given objective. Assuming we have a budget, we can add or remove noise on our training data to simulate this problem, and using reverse-mode differentiation come up with an optimal setting for this noise. With appropriate noise, we can show that this budget is best over alternatives when the noise is truly the result of aggregation from statistics.