Final Report Summary - REVVAC (Identification of vaccine candidates using reverse vaccinology)
The objectives of this Reverse vaccinology (RevVac) project were: (1) To improve the prediction of vaccine candidates, (2) To understand what makes a bacterial protein a good vaccine candidate, and (3) To create a vaccine resource for bacteriologists.
A collaboration with Prof. Helen McShane (University of Oxford) and Dr. Ann Rawkins (Public Health England) was forged, in order to test the protective efficacy of BPA predictions against infection with Mycobacterium tuberculosis (Mtb) in a mouse model. This collaboration focussed on predictions made by the RV classifier published by Bowman et al. (2011), on which this work built. Additional funding was secured, from a Confidence in Concept grant call from the Tropical and infectious disease consortium in the UK, for six Mtb potential vaccine candidates to by synthesised in the laboratory and tested in animal challenge models of Mtb. In the initial experiment one of these vaccine candidates achieved significant levels of protection against Mtb. Unfortunately, in repeated trials this protection was shown not to be robust.
To expand on the previous work by Bowman et al. (2011) we identified 64 new bacterial protective antigens (BPAs), implemented 13 new protein annotation tools, constructed new classifiers based on machine learning (support vector machines) to discriminate BPAs from non-antigens. We have now developed a fully nested cross-validated classifier. Another major finding was a bias in the selection of non-antigens with respect to previously published studies which has now been corrected so that they represent the same bacterial species and subcellular localization as their paired BPA. Additionally, we focussed on phase 2 of this project which intended to better describe what makes a protein a vaccine candidate. We found that the most informative feature for predicting a proteins protection was whether the protein was predicted to be a lipoprotein, which was also corroborated by the literature. Interestingly this identified that intra- and extracellular BPAs are fundamentally different based on the annotation derived from protein annotation tools and separate classifiers for these antigen types have been developed (Phase 1). Work from this grant has already resulted in two published manuscripts “The Promise of Reverse Vaccinology” and “Enhancing the Biological Relevance of Machine Learning Classifiers in Reverse Vaccinology”. The socio-economic impact of this work cannot be understated, especially in the current environment with levels of antibiotic resistance being concerning enough to break into common news channels as vaccinology represents a viable alternative to antibiotics for protection against bacterial pathogens.
A collaboration with Prof. Helen McShane (University of Oxford) and Dr. Ann Rawkins (Public Health England) was forged, in order to test the protective efficacy of BPA predictions against infection with Mycobacterium tuberculosis (Mtb) in a mouse model. This collaboration focussed on predictions made by the RV classifier published by Bowman et al. (2011), on which this work built. Additional funding was secured, from a Confidence in Concept grant call from the Tropical and infectious disease consortium in the UK, for six Mtb potential vaccine candidates to by synthesised in the laboratory and tested in animal challenge models of Mtb. In the initial experiment one of these vaccine candidates achieved significant levels of protection against Mtb. Unfortunately, in repeated trials this protection was shown not to be robust.
To expand on the previous work by Bowman et al. (2011) we identified 64 new bacterial protective antigens (BPAs), implemented 13 new protein annotation tools, constructed new classifiers based on machine learning (support vector machines) to discriminate BPAs from non-antigens. We have now developed a fully nested cross-validated classifier. Another major finding was a bias in the selection of non-antigens with respect to previously published studies which has now been corrected so that they represent the same bacterial species and subcellular localization as their paired BPA. Additionally, we focussed on phase 2 of this project which intended to better describe what makes a protein a vaccine candidate. We found that the most informative feature for predicting a proteins protection was whether the protein was predicted to be a lipoprotein, which was also corroborated by the literature. Interestingly this identified that intra- and extracellular BPAs are fundamentally different based on the annotation derived from protein annotation tools and separate classifiers for these antigen types have been developed (Phase 1). Work from this grant has already resulted in two published manuscripts “The Promise of Reverse Vaccinology” and “Enhancing the Biological Relevance of Machine Learning Classifiers in Reverse Vaccinology”. The socio-economic impact of this work cannot be understated, especially in the current environment with levels of antibiotic resistance being concerning enough to break into common news channels as vaccinology represents a viable alternative to antibiotics for protection against bacterial pathogens.