We propose to develop a computational approach for completely automated prediction of biochemical functions for large sets of protein-coding sequences. Instead of the conventional approaches that rely either on one round of database search and annotation inheritance from the best-scoring sequence match, or on manual case-by-case reinspection, we will construct proprietary structured database of complete proteomes (Proteome Bank) and develop automated multi-step strategy of database searches, feature prediction, and annotation. The cascade of analysis will include: automated selection of reference databases (Proteome Bank, other proprietary database; public database, etc.), several types of filtering, domain dissection, iterative search, and rule-based modification of the database annotations. The system will analyze one or more proteins every five minutes, which is a ten fold increase over the productivity of a highly trained analyst. In addition, custom heuristic algorithms should result in 20-30 percent greater accuracy in annotations as compared to the existing automated approaches. The ability to predict protein functions rapidly and accurately at the genome scale will be applied for exhaustive annotation of proteomes and EST collections from humans, animals, plants and microbes, in order to reconstruct biochemical pathways and prioritize them as targets for modification.
The genome-scale functional annotation of protein sequences at high speed and accuracy will accelerate gene discovery, target validation and metabolism reconstruction. This will provide for rapid and reliable development of human therapeutic molecules, antimicrobials, agricultural products, and industrial enzymes. Many new functional predictions for proteins from different species will be made and value-added sequence databases will become available.