NERVE: new enhanced reverse vaccinology environment

Abstract

Background: Since a milestone work on Neisseria meningitidis B, Reverse Vaccinology has strongly enhanced the identification of vaccine candidates by replacing several experimental tasks using in silico prediction steps. These steps have allowed scientists to face the selection of antigens from the predicted proteome of pathogens, for which cell culture is difficult or impossible, saving time and money. However, this good example of bioinformatics-driven immunology can be further developed by improving in silico steps and implementing biologist-friendly tools.

Results: We introduce NERVE (New Enhanced Reverse Vaccinology Environment), an user-friendly software environment for the in silico identification of the best vaccine candidates from whole proteomes of bacterial pathogens. The software integrates multiple robust and well-known algorithms for protein analysis and comparison. Vaccine candidates are ranked and presented in a html table showing relevant information and links to corresponding primary data. Information concerning all proteins of the analyzed proteome is not deleted along selection steps but rather flows into an SQL database for further mining and analyses.

Conclusion: After learning from recent years' works in this field and analysing a large dataset, NERVE has been implemented and tuned as the first available tool able to rank a restricted pool (approximately 8-9% of the whole proteome) of vaccine candidates and to show high recall (approximately 75-80%) of known protective antigens. These vaccine candidates are required to be "safe" (taking into account autoimmunity risk) and "easy" for further experimental, high-throughput screening (avoiding possibly not soluble antigens). NERVE is expected to help save time and money in vaccine design and is available as an additional file with this manuscript; updated versions will be available at http://www.bio.unipd.it/molbinfo.

Figure 1

NERVE software pipeline. The process can be divided into two parts: data production and storage (top) and data selection (bottom). Six different scripts screen the entire proteome to mine and infer information that flows into a MySQL table. A seventh script uses four filters (LOC, localization; TOP topology; PAD, probability of being adhesin; SHP, similarity to human proteins) and analyzes values created by steps 1 through 5 to select and rank VCs that are then presented in a html table with links to relevant data.

Figure 2

Flow-chart of NERVE working process. Amino acid sequences from the whole bacterial proteome undergo six analytical steps: prediction of subcellular localization (1), calculation of probability of being adhesin (2), identification of TM domains (3), comparison to the proteome of Homo sapiens (4) and to that of a pathogen selected by the user (5), assignment of a putative function (6). Each of these steps stores data mined in an SQL database. After filtering and ranking, the best VCs are presented in a user-friendly html table (see figure 1 and Results and Discussion for details).

Figure 3

Data concerning the ten proteomes used for tuning NERVE. The number of selected VCs is reported beside the overall number of sequences. The average size of the selected VC pools is 8.17% of the proteome (min 5.09%, max 10.73%).

Figure 4

Data concerning the six proteomes used to test NERVE settings. The number of selected VCs is reported beside the overall number of sequences. Average size of selected VCs pools is 9.32% of proteome (min 8,17%, max 11,33%).