We try to predict aspects of protein structure and function from the evolutionary information contained in families of protein sequences. The main current projects are to (1) predict functional classes for orphans, (2) the sub-cellular localisation from sequence, and (3) protein-protein, and protein-substrate interactions from sequence and predicted structure. We also continue to work on structure prediction. Particular goals are: (1) Improve predictions for secondary structure, solvent accessibility, transmembrane proteins, (2) develop methods predicting residue-residue distances, and (3) improve methods that allow to model the 3D structure for proteins that have distant homologues of known structure (threading). Additionally, we work on describing and clustering the space of protein sequences and structures in context of structural genomics. Recently, we have analysed proteins with non- regular structures in detail.
Both in order to accomplish our goals in bioinformatics and to make our methods available, we maintain local copies of most relevant sequence databases. We also maintain a variety of servers that assist biologists in their every day sequence analysis: PredictProtein, META-PP, EVA, PredictNLS. Finally, we are working on a database that collects predictions and alignments for all entirely sequenced organisms. The technical means we apply range from scanning literature, over simple statistical methods, to methods from artificial intelligence (neural networks, hidden Markov models, or genetic algorithms).