The junior research group Data Inspection develops and studies novel computer algorithms for the analysis of biological high-throughput data.

The central goal of biological experiments is the collection of new and meaningful data. Measurements do often reflect a condensate of laborious and costly work concerning experimental design, setup, and execution. Since data analysis is the key to gain information and further knowledge about the subject of interest, faithful data inspection methods are required to extract substantial facts from the measurements. 

In addition to the analysis of existing data sets, data inspection aims at creating auxiliary data models that integrate prior knowledge, such as contrast information, class labels, or even only loosely associated observations. Models of this kind include Markov models, such as hidden Markov models (HMM) or Markov random fields (MRF), and extensions of prototype models, such as self-organizing maps (SOM), neural gas (NG), and learning vector quantization (LVQ).

These models are data-driven, i.e. the data space induces a specialized model space for facilitating and focusing the analysis. Problem-specific metric adaptation is one particular and powerful case of this concept, allowing feature rating and feature selection for biomarker detection, as well as improved clustering and classification. Other modeling targets are motif detection in directed data, trustworthy data visualization, and alternative data views that help to overcome limitations of standard statistical methods.

In order to meet the interests of our cooperation partners, the research emphasis is put on the processing of genomic sequences, macroarray and microarray data, and large gelplot collections. Being data-driven, though, the models allow a very broad range of biological applications.

The research group is funded by the Ministery of Culture of Saxony-Anhalt, grant XP 3624HP/0606T.