LWA 2007: Lernen - Wissen - Adaption, Halle, September 2007, Workshop Proceedings

Due to advances in experimental techniques wehave seen a vast increase in exciting new datareflecting the spatio-temporal aspects of life onthe molecular level. It augments static informa-tion about DNA sequences, transcripts and pro-teins and their interactions. The integration ofthese heterogeneous sources of data in order toarrive at conclusive information about a biologi-cal process is typically performed manually. We revisit classical mixture models, or convexcombinations of density functions, and showhow they can effectively model high-dimensionaldata by use of sufficiently constrained compo-nent models, for example for gene expressiontime-courses. One approach to combining abun-dant primary data with labels from possiblysparse secondary data is semi-supervised learn-ing. We will use the detection of groups of syn-expressed genes from in-situ images and gene ex-pression time-courses during embryogenesis us-ing a semi-supervised approach as a case study.INVITED TALK