International Workshop on
Probabilistic Modelling in Computational Biology 26 July 2006 Vienna

The extraction of an improved biological understanding from genome-scale quantitative measurements, like microarray data, has recently been a very active area of bioinformatics research. In particular, models based on these data are used in the investigation of complex interactions with many players, many of which cannot easily be studied in traditional ways because experimental interference is either lethal or shows no obvious phenotypical effect due to redundancies. Frequent tasks in such research work include the identification of common molecular mechanisms and the detection of causal effects, in particular, the mapping of regulatory pathways.

The required inference of the often highly complex models of biological processes involves large numbers of interacting variables that have to be chosen from a pool of tens of thousands of candidates. The high dimensionality of the problem is starkly contrasted by the very small numbers of samples that can be obtained by direct assays. This makes an efficient use of contextual information critical.

The difficulty of an integrated analysis is compounded by the heterogeneous nature of contextual knowledge available. Data sources vary considerably with respect to comprehensiveness, error rates, and suitability for direct computational exploitation. Besides data from microarrays for gene expression profiling, some typical representatives of different data classes are: abstracted relationships between biological entities (e.g., Gene Ontology); curated pathway descriptions (e.g., KEGG or EcoCyc), curated functional annotation (e.g., UniProt/SWISS-PROT); results from de-novo sequence analysis (e.g., phosphorylation site predictions); data from protein-protein interaction experiments (e.g., Y2H screens); and experimentally detected protein-DNA binding sites (e.g., ChIP-chip results).

Efficiently taking advantage of these data raises many technical, conceptual, and analytical challenges. During the workshop, which this year is officially affiliated with the ISMB/ECCB 2007 conference in Vienna, we will explore these issues, with an emphasis on probabilistic methods in three themed sessions.