The bacterial virulent protein sequences were retrieved from the SWISS-PROT [20 (link)] and VFDB (an integrated and comprehensive database of virulence factors of bacterial pathogens, [21 (link)]). SWISS-PROT sequences were retrieved using keywords such as virulence, adhesin, adhesion, adherence, toxin, invasion, capsule and other terms related to virulence factors. The VFDB and SWISS-PROT sequences were screened strictly in order to obtain a high quality dataset. First, the sequences were filtered to remove entries annotated as "Probable", Putative", "By similarity", "Fragments" "Hypothetical", "Unknown" and "Possible". The filtering yielded 1756 annotated virulent protein sequences (henceforth referred to as positive dataset).
For training with non-virulent protein sequences, we selected 3000 annotated protein sequences of bacterial enzymes and other non-virulent proteins from SWISS-PROT database (these sequences are henceforth referred to as negative dataset). The negative dataset sequences were mainly chosen from the bacterial proteomes, the virulent protein sequences of which are included in the positive dataset.
For training with non-virulent protein sequences, we selected 3000 annotated protein sequences of bacterial enzymes and other non-virulent proteins from SWISS-PROT database (these sequences are henceforth referred to as negative dataset). The negative dataset sequences were mainly chosen from the bacterial proteomes, the virulent protein sequences of which are included in the positive dataset.
Free full text: Click here
