Georgetown University Medical Center Washington United States
Biomedical literature represents the primary source of experimental data and biological knowledge. This project developed a text mining system for pathogens of biodefense relevance, focusing on mining pathogen-host proteomic data. We developed a Support Vector Machine SVM-based system to identify abstracts containing protein interaction information using an annotated corpus of 1360 MEDLINE abstracts as the training set. It achieved good performance on document classification with a precision of over 80 among top 50 ranked abstracts. The SVM-based method is further augmented with other text mining tools such as PIE for mining and tagging PPI information. As part of an effort in enabling text mining tools for real world applications, we coupled our analysis with the functional annotation of proteomic experiment. All the data was then loaded into iProXpress system and provided to the collaborating USAMRIID laboratory for the analysis of bacterial pathogen proteomics data.