The sequencing of many eukaryotic, bacterial, archaeal and viral genomes during the last decade has revolutionized our understanding of biology, ecology and evolution. The exponentially growing number of genomic and metagenomic sequences demanded the implementation of fully automated software systems for genome annotation. Due to very high computational costs, the calculation of sequence similarities and detection of protein domains limits the processing and update performance in genome annotation. Therefore we have developed the Similarity Matrix of Proteins (SIMAP) database, providing a comprehensive and up-to-date dataset of a pre-calculated sequence similarity matrix and sequence-based features like InterPro domains for all proteins contained in the major public sequence databases. In order to structure the protein sequence space, SIMAP provides an integrated clustering based on sequence similarities and domain architectures. Mapping Gene Ontology Annotations (GOA) of known proteins to these clusters provides reasonable protein function predictions for large parts of the protein sequence space. SIMAP accelerates several important bioinformatics resources, as PFAM, Gene3D, PEDANT, MEGAN and Blast2GO. In addition to large-scale computational approaches as SIMAP, the computational interpretation of genomes has to keep up with the emerging knowledge about the molecular basis of specific biological processes. Our recent development of sequence based prediction methods of Type III and Type IV secreted effector proteins has addressed two key mechanisms for infection, pathogenesis and modulation of infected hosts by pathogenic bacteria. Although a number of expensive genome wide screens for novel effector proteins have been performed, no computational model had been published for the general de novo prediction of Type III and Type IV secreted proteins. Based on comprehensive and manually curated databases of known effectors, we discovered sequence signals that are typical for these proteins. Modeling of the signals using a machine learning approach enabled genome-wide predictions of Type III and Type IV secretomes in many bacterial genomes and will contribute to the development of novel, specific antibiotics.