A study in npj systems biology and applications in April 2024 presents an approach for “identifying and prioritising vaccine antigens” using a machine learning-based reverse vaccinology strategy to predict potential new malaria vaccine candidate antigens. The team behind the paper comprises computational biologists at the University of Maryland, College Park, and researchers at the University of Maryland School of Medicine in Baltimore. Not only did they identify new potential targets but were able to rank them based on performance.
A malaria need
The authors recognise “substantial reductions in the malaria burden in many endemic areas” thanks to artemisinin-based combination therapies and other tools.
“However, progress toward malaria elimination has stalled as malaria incidence has plateaued and gains have been threatened by the emergence of resistance to interventions in the parasite and vector.”
The role of malaria vaccines in elimination efforts will be “critical”. Transmission of Plasmodium parasites occurs after infective mosquitoes take a blood meal and inject sporozoites, which then develop and multiply in the liver. Vaccines are “meant to block infection” against this pre-erythrocytic stage. Once the Plasmodium merozoites emerge from the liver they “invade and replicate inside red blood cells”. This stage of the life cycle causes malaria disease and death, which blood-stage vaccines are intended to “limit”. Vaccines that block transmission would “inhibit parasite sexual reproduction and development in the mosquito”.
“Design of a broadly protective malaria vaccine has been hampered by several factors, including multiple parasite life stages that express different antigens, extensive genetic diversity within individual antigens targeted by vaccines, partial natural immunity that is short-lived and non-sterilising, and incomplete knowledge of immune correlates of protection.”
Malaria parasites, haploid in humans and “briefly” diploid in mosquitoes, have extensive genetic variation “generated through mutation during mitotic reproduction in humans and by sexual recombination in the mosquito”. Although the first P. falciparum genome was shared in 2002, most vaccine development efforts have focused on a “small number of highly diverse vaccine candidates”. These have been identified through “traditional” vaccinology approaches, in contrast to a “more comprehensive, genome-level approach”.
Reverse vaccinology
A strategy first proposed by Dr Rino Rappuoli is reverse vaccinology, which uses bioinformatics approaches to identify pathogen antigens or epitopes that can serve as vaccine candidates. It has been used in the identification of several vaccine antigens for bacterial and viral pathogens.
“The wealth of systems data available for P. falciparum lends itself to the use of reverse vaccinology to identify new malaria vaccine antigens, which may allow identification of less immunodominant but more conserved antigens that have been misused using traditional vaccinology approaches.”
This, the authors suggest, is where machine learning will be useful; it doesn’t require a priori assumptions about the importance of specific criteria, but “learns” protein properties most associated with vaccine potential. Positive-unlabelled (PU) learning identifies potential positives among unlabelled entities “based on the properties of the positives”. It has been used to identify genes associated with human disease but not, as far as the authors are concerned, to identify candidate antigens.
In their study, the authors modify canonical positive-unlabelled random forest (PURF) to distinguish proteins with vaccine potential (antigens) from non-antigens, ranking candidates with probability scores.
What does the study find?
The authors chose random forest because of its “high predictive accuracy, high interpretability, and insensitivity to outliers, and predictive variable scales”. The approach identified “previously unknown vaccine candidate antigens” within a “flexible framework” that allows prioritisation. Candidate antigens were filtered based on gene essentiality, where mutations could affect parasite viability and help reduce parasite escape from vaccine-induced immune responses. Many of the candidate antigens were “predominantly” expressed in a single life stage, and those that were expressed in multiple life stages may be “attractive” because they would target multiple life stages. A summary report of the candidate antigens can be viewed here.
The approach is not limited malaria; the authors believe that the methodology can be “tailored and applied to other disease pathogens” and could inspire other scientific research areas. Professor Michael Cummings, one of the co-authors, commented that “these findings offer a flexible framework for future vaccine research”.
“We can adjust our criteria and even apply this approach to other diseases beyond malaria. It’s a big step forward in the quest for better vaccines.”
For the latest vaccine technology updates and insights into infectious disease management, why not join us at the Congress in Barcelona or subscribe to our newsletters here?
