Antibacterial Drug Target Candidates Discovery System

Infectious diseases become a growing threat to public health due to the antibiotic resistance in old pathogens and the emergence of new pathogens. Novel antibiotics against the bacterial pathogens are expected. Genome projects provide a new strategy for antibiotics development. This thesis reports an antibacterial drug target candidates discovery system, based on bioinformatic analysis on all fully sequenced pathogens.We have developed four bioinformatics tools, which were integrated into the system. First, we built up a distributed BLAST system that is a loose coupling and heterogeneous parallel system parallelizing the BLAST programs on a task level. The system is powerful, economical, well portable and scalable. Second, we designed a linear discriminant model, using iterative discriminant analysis method to integrate three DNA composition criteria (GC difference, dinucleotide relative abundance difference, codon usage bias) and taking known pathogenicity islands as training set. This model can be used in detecting pathogenicity islands and other anomalous gene clusters. Next, we collected virulence factors from literatures and databases, constructed a virulence factor database, which can be used in identifying the related factors in bacterial genomes. Finally, we developed a literature mining software called MedBlast to search articles related to a biological sequence. Preliminary test showed that the average recall rate of this program reaches 75.1%.Based on the above methods and various other techniques, we constructed the antibacterial drug target candidates discovery system. This system was applied to three different data sets employing different information: pathogenicity, conservation and localization. First, it is well known that pathogenicity related genes in pathogens are potential drug targets. We analyzed all pathogen genomes by the iterative discriminant analysis method to detect pathogenicity islands; then we identified the potential virulence factors with the virulence factor database. These two methods are based on composition analysis and homology searching respectively, which can be complement each other and achieve better performance in detecting pathogenicity related genes. Second, conserved genes among many genomes are also good candidates for drug target, because they are likely to be essential genes, and good targets for broad-spectrum antibiotics. We used ortholog gene groups to identify the conserved genes in Clusters of Orthologous Groups and KEGG Orthology databases. Third, we used bioinformatics techniques to identify outer membrane proteins and lipoproteins, which are often easy targets for drug and vaccine because of their accessibility. Next, all the identified genes were subtracted against human proteome. The bacterial proteins that are homologous to human proteins are discarded to avoid toxicity. Finally we annotated the targets with MedBlast and Gene Ontology Annotation database to facilitate further research. Above all, using database and web technology, we built up the antibacterial drug target candidates discovery system.The system is innovative. In the test on H.pylori 26695 we discovered many potential antibiotic drug targets efficiently, including house keeping proteins, pathogenicity islands, toxin, drug resistance proteins, outer membrane proteins, iron uptake proteins, which showed that it is a valuable bioinformatics analysis system. It can be improved in several aspects. Combined with the experimental research, it will be helpful for the development of novel antibiotics and improving human health.