PP-042 Comparison of three methods for the detection of biofilm forming microorganisms isolated from a tertiary care hospital in Pakistan

A. Hassan, J. Usman, F. Kaleem
2010 International Journal of Infectious Diseases  
Poster Presentations S37 Result: Out of 267 samples, 133 gave positive results for B. abortus by real-time PCR. Conclusion: These results indicate a high presence of this pathogen in this area of the country and that this method is considerably faster than current standard methods. PP-041 Computational prediction of type III secreted proteins using labeled and unlabeled data Background: The type III secretion system (T3SS) is a specialized protein delivery system that injects proteins
more » ... ) directly into the eukaryotic host cytosol and facilitates bacterial infection. T3SS plays a key role in pathogens, but its secretion mechanism has not been fully understood yet and the type III secreted effectors (T3SE) are notoriously difficult to identify. We propose to predict T3SEs with a bioinformatics method since the wet-bench approaches, e.g., functional screen and protein secretion assay, are laborious and timeconsuming. Moreover, because the confirmed effectors and non-effectors are very few, our method is based on semisupervised learning utilizing both labeled and unlabeled data to improve the prediction accuracy. Methods: We adopted SVMlin as the predictor. It implements linear SVMs and also extensions of standard SVMs to incorporate unlabeled data. The feature vectors involve amino acid composition, secondary structure and solvent accessibility information. Result: We have built a non-redundant data set from Pseudomonas syringae, including 108 positive samples and 3424 negative samples. This data set was divided into five subsets, four of which for training and the left one for test. In addition, 3000 unlabeled data were used in the semi-supervised learning. The results are listed in Table 1 . Three measures were examined. We observe that using the unlabeled data helps to improve the recall nearly 20%. Although the precision decreases, the total accuracy is still improved. Overall, the semi-supervised learning has a better performance.
doi:10.1016/s1201-9712(10)60110-5 fatcat:kv53hg5imnb7hov3ar4gegtdj4