Shortly after the sequencing of the first bacterial genome, Rino Rappuoli proposed the original hypothesis of in silico antigen discovery, leveraging genomic information for vaccine design in the late 1990s (1). In 2007, he provided proof-of-concept to support this, with a RV-driven trivalent formulation that protects against Neisseria meningitidis group B (2), licensed as Bexsero (3, 4). This highlighted that RV was not only a pure ideation but a useful approach that generates true biomedical value.

RV has had considerable promise as an alternative to traditional experimental screening of antigens, offering benefits in cost, time, effort, and safety. Immunoinformatics has substantially reshaped the way antigens are identified through the prediction of B-cell and T-cell epitopes at an unprecedented speed and scale (5). The RV paradigm has been progressively enriched with the incorporation of novel filtering criteria—such as gene essentiality, high sequence conservation, subcellular localization, and predicted off-target effects—and the refinement of pre-existing ones (6). Five technical pillars have further promoted RV, namely, (i) availability of many fully-sequenced genomes for most virulent microorganisms; (ii) three-dimensional information for nearly all proteins (7); (iii) the arousal of a large community of algorithm developers; (iv) the production, standardization and storage of immunological data in accessible resources; and (v) the universalization of supercomputing. As a result, RV pipelines have populated the field of fundamental antigen discovery for more than two decades (8). Consequently, they have been applied to virtually all relevant bacterial and viral pathogens, and extended to some fungi and parasites.

Despite the undeniable contributions of RV to the theoretical vaccine field, the number of licensed immunoprophylactic products that have directly resulted from this strategy remains disproportionately low. This mismatch between hypothetical and practical success has raised concerns across the field. For instance, many RV-predicted antigens fail to induce protective responses in vivo, revealing persistent challenges such as low predictive power for conformational epitopes and insufficient capacity to provide immunological context to data (9–11). Furthermore, virtual vaccine design may not have achieved critical data and algorithmic maturity to overcome constraints encountered when modeling the complexity of the desired response to elicit (12). Altogether, these drawbacks affecting RV can misguide research efforts and resource allocation.

In this study, we examine the current state of RV, reflecting on its achievements, limitations, and points of stagnation. Also, we have interrogated the relevance of RV publications besides the molecular nature of the resulting antigens and the selected pathogen types, and interpretated the trends of the outcomes. In addition, we have explored the nature of antigen selection—and design—of current approved vaccines or candidates under evaluation in human trials, and the role of RV in the principal human infectious threats. Finally, we propose potential lines of evolution to guide the next generation efforts of computational vaccinology, including potential meeting points between RV and late vaccine development players.

Antigen selection via computer simulations emerged almost three decades ago as a postgenomic tool suited for vaccine development. However, this has not delivered a substantial portfolio of effective immunological solutions against the most pressing infectious threats. Thus, RV-like techniques run the risk of being relegated by vaccine experts to purely speculative tools. The misalignment between RV and the development of vaccine end products appears multicausal and due to internal and external reasons. While the former ones mostly relate to limited performance on experimental tests, the later ones are associated with a lack of familiarity among late-stage vaccine developers and a scarcity of computationally emergent precedents. Here, we have attempted to audit the applicability of the RV idea over time, current fundamental limitations and possible strategies to deal with them.

Although secondary reasons may also be involved (e.g. epidemic or editorial board scope changes), an analysis of the explosion of RV reports from 2021 onwards strongly suggests growing weaknesses. Although arguable, these weaknesses very likely hamper the performance of the field and are responsible for the progressive decline in the impact of RV papers and growing skepticism in some areas. These include the recurrent absence of experimental verification, the steady movement to debatable multi-epitope hybrid solutions and the selection of target pathogens with already optimal public health options or with those that hardly will return the investment. Unless RV confront real gaps in vaccine availability, i.e. pathogens with large impact on the population for which there is no vaccines with broad and durable protection, several concerns will persist. These concerns relate to redundancy and the rationale for research for replacing well-established and effective decades-old vaccines in pandemics, multidrug resistance and calendar pathogens.

Mature preclinical vaccine antigen candidates and end products must be tightly adapted to the pathoimmunology of the germs and the expected target population. This requires multidisciplinary expert teams and expensive facilities to obtain key experimental support (39). As a gold illustrative example of RV efficacy, Bexsero development was based on multiple antigen expression in E. coli, rodent immunization, and immunological assays. In contrast, most RV cases converged into comparable report-like procedures executed using universal algorithms and data—in some cases without a deep immunological or microbiological experience regarding the pathogen in question—and no functional validation. Thus, the promise of cheap and rapid investigations, can make the RV community to succumb to the temptation of short-term publications, avoiding the risk of costly laborious experiments. However, this is at the expense of generating preliminary results with unknown prophylactic value, compromising the ultimate goal of vaccine development. Thus, RV history missed the opportunity to refine original predictions and distil causal insights, at least, by successive iterations with in vitro test or preclinical models. Understandably, simplicity and low cost could contribute to the democratization of early-stage vaccine discovery in environments with limited access to advanced experimental installations. In these cases, international collaboration would facilitate coupling with appropriate validation pipelines potentially uncovering antigens that would otherwise remain unexplored.

Advantages such as design flexibility and immunogenicity breadth explain the RV community drift towards multi-epitope constructs. However, these protocols present important concerns that reduce their biotechnological attractiveness even today (40). The expected improvements resulting from combining epitopes from several antigens are deeply dependent of rules regarding epitope-centered protection, inter-epitope synergy and immunodominance that are far from being understood (41). In contrast, Bexsero, the epitome of successful RV, includes three complete antigen proteins. Admittedly, multi-epitope vaccines warrant a renewed experimental exploration but, to our knowledge, none advanced to further human stages and very few studies involved animal models (42). Thus, the overrepresentation of hybrid schemes with insufficient support in the RV field adds considerable uncertainty to the area and likely contributes to the disconnect between RV scientists and vaccine production lines.

RV is one among a handful of theoretical techniques that perceive rational vaccine design from different angles (43). Several emergent fields of study marginally overlap with RV but also complement it as, for instance, systems vaccinology approaches the host response of the vaccinees in a multi-omic manner (44); network vaccinology aims to model the complex interactions between immune system elements (cells, proteins, metabolites, etc.) using the strength of graph theory (45); and structural vaccinology takes advantage of three-dimensional structures to optimize epitope analysis of antigens (46). Despite the inclusion in RV of the term “vaccinology” (defined as “the science and study of vaccines, encompassing their development, production, immune system response to them, safety, etc.”), in practice RV remains focused on the initial selection and/or design of antigen candidates.

Thus, to reconcile semantics and fully contribute to vaccine development, we believe that RV should be integrated at least with the complementary approaches indicated above. Ideally, this should be carried out under the guidance of multidisciplinary experts and encompassed within a more holistic framework such as “Knowledge-based vaccinology” (KBV). KBV tools should aim to identify optimal combinations of antigens, adjuvants, delivery platforms, dosages, vehicles, and administration regimens, as this is experimentally unapproachable due to the curse of dimensionality (47). KBV should establish intellectual and computational bridges among these categories to enhance their performance beyond what can be achieved when they are considered in isolation. We anticipate an approach analogous to personalized oncology, in which the convergence of genetics, pharmacology and machine learning enables the prediction of effective treatments (48). As the fusion of the involved disciplines is still in its infancy, it is difficult to provide concrete examples of advanced KBV prototypes. Thus, we envisage the KBV roadmap will evolve through a gradual transitional phase to become the standard for rational vaccine design, in which classical RV and higher-order approaches will coexist, rather than resulting from an abrupt conceptual leap.

External factors affecting RV include the tendency of large pharmaceutical companies and public entities to perpetuate successful past antigen types. Attenuated/inactivated microorganisms, polysaccharidic capsular components, in addition to viral spike-like proteins, abundant outer proteins and bacterial fimbrial adhesins nearly monopolize licensed vaccines or those under study in humans. Despite the apparent reductionism, reluctance to utilize non-classical antigenic strategies is grounded in the pressing need at scale to ensure consistent performance and minimize legal, safety, and public cost risks. This clashes with theoretical scientific and publishing pressures of pure RV teams resulting in a hypertrophy of the number of vaccine candidates opening the gap with respect to those probed useful. Consequently, the RV community’s lack of integration into the complete vaccine lifecycle creates a conflict between early-stage and late-stage developers.

To overcome this barrier, the RV groups need to participate and reach a critical mass of real-world antecedents. This trajectory would persuade large companies and public institutions to invest in their outcomes. To balance both worlds, bidirectional awareness initiatives, targeted meetings, and interdisciplinary consortia could emphasize the potential of data science and establish new lines of thinking among project evaluators, R&D directors at pharmaceutical firms and policymakers. In return, these key stakeholders should provide insight concerning epidemiology, social expenses, society damage, legal background, large-scale calculations and organize human cohorts following refined RV suggestions. This would forge a solid generation of vaccine experts that unify the RV strengthens and late challenges. Otherwise, attachment to pre-genomic paradigms surely make governments miss optimal opportunities that remain in the shade and RV initiatives will “die on the shore” without human-testing.

Overall, we suggest RV must extend its focus to experimental validation, encompass the entire vaccine lifecycle and foster closer collaboration with late-stage vaccine development stakeholders. Otherwise, despite its contributions to early antigen selection, RV may be limited in delivering on its promise of transforming the vaccine development pipeline. In any case, we remain optimistic that the knowledge-based approach to vaccine design initiated with RV is merely undergoing growth pains and still has the potential to evolve into a powerful framework to address urgent infectious threats.