In the following, the single quotation marks ‘’ are used to indicate an ontology term. The italics are used to refer to an ontology relation terms (i.e. object properties).

Cancer vaccine data

A total of 677 cancer vaccines targeting 42 cancer types, along with associated data, were collected in the CanVaxKB database. To ensure consistent representation of cancer vaccine data, we mapped the vaccines, vaccination routes, and vaccine platforms to VO, vaccine organisms and pathogens to NCBITaxon, and cancer types to the Human Disease Ontology (DOID) [36, 37] and the Mondo Disease Ontology (MONDO) [38] when the terms are not available in DOID.

Of the 42 cancer types, 39 were found in the DOID. For the three missing cancer types, two were identified in MONDO and we imported terms aligned with DOID in VO. One cancer type, ‘HPV associated cancer’ was not available in any OBO ontologies. ‘HPV associated cancer’ was present in VO as “A cancer that is caused by human papillomavirus (HPV) infection. HPV infection can lead to six types of cancer including anal, cervical, oropharyngeal, penile, vaginal, and vulvar cancer.” with logical axioms:

‘HPV associated cancer’ = (equivalent to: ) cancer and ‘disease has basis in’ some ‘HPV infection’.

Five vaccination routes and four vaccine platform types associated with cancer vaccines have been defined in VO. All vaccine organism and pathogen types were available in the NCBITaxon. However, only a few cancer vaccines existed in VO. Most of them need to be added to VO.

Ontological design pattern for cancer vaccines

Traditionally, vaccines are developed to prevent or treat infectious diseases caused by pathogens such as bacteria, viruses, and protozoa [39]. Recently, there have been increasing studies in developing vaccines for diseases not typically caused by pathogens, such as cancer, allergy, and autoimmune diseases. To model cancer vaccines, we initially defined the term, ‘cancer vaccine’ in VO, as a vaccine that prevents or treats cancer. Logically, ‘cancer vaccine’ is equivalent to:

‘vaccine’ and ‘immunizes against disease’ some ‘cancer’.

This implies that any vaccine capable of immunizing against cancer is classified as a cancer vaccine.

In ontology, shortcut relations are often used to simplify and streamline complex relations among entities, making the ontology easier to understand and use. These shortcut relations are usually defined and only accessible in the ontology; however, their usages are often important for simplification, enhanced clarity, improved usability, and efficiency. In the VO, three key shortcut relations have been defined to link vaccines to critical entities associated with vaccines:

‘immunizes against disease’: a shortcut relation between vaccine and disease (e.g., cancer) that the vaccine can prevent or treat.

‘immunizes host’: a shortcut relation between vaccine and the organism that assumes the ‘vaccine host role’ is the target of a vaccine administration.

‘immunizes against pathogen’: a shortcut relation between a vaccine and a pathogen wherein the vaccine targets against the pathogen in the immunization process.

We adopted the existing vaccine design pattern and extended it to create the cancer vaccine specific ontology design pattern (ODP) shown in Fig. 1. Two object properties ‘immunizes against disease’ and ‘immunizes host’ are used to represent the relations between a vaccine and a disease and a host, respectively. The relation ‘immunizes host’, a shortcut linking a vaccine and the immunized organism assuming the ‘vaccine host role’. However, ‘vaccine host role’ in VO is a subClass of ‘host role’ from the Ontology for Biomedical Investigations (OBI) [40]. In OBI, ‘host role’ is defined as:

Fig. 1

Cancer vaccine ontology design pattern (ODP). The pattern represented the collected cancer vaccine data (shown in box as VO classes) and the relations (as VO object properties) between them

‘host role’ = def: A role played by an organism that provides nourishment, shelter, or a means of reproduction to another organism within the organism.

The above ‘host role’ only applies for organisms such as bacteria and protozoa. However, such a ‘host role’ may not be correct for cancer vaccines since most cancer vaccines do not deal with two organisms as seen in the ‘host role’ definition. To address this issue, we redefine the ‘vaccine host role’ in VO under the BFO: role (instead of under ‘host role’) specifically for the host of vaccine. Note that unlike the ‘host role’ definition provided above, the host target in our new term is vaccine, a foreign material entity that does not have to be another organism. The VO term ‘vaccine host role’ is now defined as follows:

‘vaccine host role’ = def: A role that inheres in an organism that hosts a vaccine being administered into the organism.

Different from traditional infectious disease vaccines, the relation ‘immunizes against pathogen’ does not apply to most cancer vaccines, as pathogens do not directly cause cancer. However, some virus infections can lead to cancer. For example, HPV infection can cause cervical cancer and other five types of cancer including anal cancer, oropharyngeal cancer, penile cancer, vaginal cancer, and vulvar cancer [10]. Consequently, anti-virus infection cancer vaccines have been developed to immunize against virus infection, thereby preventing associated cancer. For example, HPV cancer vaccine targets HPV to prevent the cancer caused by HPV infection [9,10,11].

Therapeutic cancer vaccines primarily target tumor antigens to elicit recognition and elimination of tumor cells by inducing the host immune response. For example, ‘Carcinoembryonic Antigen Peptide-1 Vaccine’ is a cancer vaccine that contains carcinoembryonic antigen, a tumor antigen, obtained from cancer cells which stimulates an immune response against tumor cells [41].

Figure 1 (bottom right) represents the ontological model of both preventive and therapeutic cancer vaccines. We defined specific ontology terms for these cancer vaccine in VO as follows:

‘preventive cancer vaccine’ = def: A cancer vaccine that prevents cancer development associated with viral infections.

‘therapeutic cancer vaccine’ = def: A cancer vaccine that aims to eliminate or control tumor cells by recognizing the tumor cells and stimulating the immune system via tumor antigens.

‘tumor antigen’ = def: An antigen expressed by the tumor cells, which may be exclusively present on tumor cells or overexpressed on them.

Further details on tumor antigen-related description and modeling are provided in the subsequent section.

Other features, such as vaccine platforms, vaccine administration routes, vaccine development status, associated with cancer vaccines are also represented in Fig. 1 (below). The models of vaccine administration routes and platforms are adopted from traditional pathogen vaccines [32]. A vaccine platform refers to the underlying technology or approach used to develop vaccines. Different vaccine platforms employ distinct strategies to stimulate the immune system and induce immunization against diseases. In VO, we modeled a platform based on the characteristics of primary material used in a vaccine. For instance, for vaccines containing pathogenic organisms, ‘organismal quality’, such as ‘organism inactivated’, ‘vaccine organism live attenuated’, describes the quality of organisms used. Otherwise, a specific ‘vaccine role’, such as ‘DNA vaccine role’ and ‘RNA vaccine role’, are used to differentiate vaccine prepared material (e.g. DNA, RNA). The ‘vaccine role’ is acquired in the vaccine preparation process and then realized in vaccine immunization. This model enables automated vaccine classification based on the platform.

Figure 1 illustrates the comprehensive process of ‘vaccine immunization’, wherein a ‘cancer vaccine’ is administered to an organism through a specified ‘route of administration’. This introduction stimulates the organism’s immune system, resulting in a ‘vaccine-induced host immune response’, and contributing to the prevention or treatment of cancer.

Ontology modeling of cancer vaccine antigens and canvaxgens

Tumor antigens play a crucial role in stimulating therapeutic immune responses against cancers. Various types of cancer vaccine antigens exist, categorized based on the target and delivery methods (see Table 1). Shared tumor antigens are common across many patients [42], while personalized tumor antigens are unique to an individual’s tumor [43]. Anonymous tumor antigens involve either In Situ antigen-presenting cell (APC) colocalized vaccines [44, 45], which colocalize antigens with antigen-presenting cells (APCs) at the tumor site, or ex vivo cell vaccines, where this process occurs outside the body [10, 46]. Additionally, dendritic cell vaccines utilize dendritic cells to present tumor antigens to the immune system [5, 47]. Each category of cancer vaccine employs distinct strategies to engage the immune system, providing a comprehensive overview of the various approaches in cancer vaccine development and cancer immunotherapy.

Table 1 Overview of various approaches in cancer immunotherapy

Efforts in cancer vaccine development focus on identifying and utilizing tumorigenic peptides as shared or personalized antigens (see Table 1) for effective vaccine development. Typically, cancer vaccines do not utilize entire genes or proteins as vaccine antigens due to the potential risks of severe adverse events, such as autoimmunity and severe diseases. Instead, cancer vaccine development often employs fragments of genes and proteins. To address the challenge of analyzing these fragments effectively without broader context, we introduce the concept of ‘canvaxgen’:

canvaxgen = def: A gene expressing a protein, either partially or entirely, serving as the antigen within a specific cancer vaccine.

It is important to distinguish canvaxgen from cancer vaccine antigen. Cancer vaccine antigens are specific proteins or peptides in cancer cells, which are targeted by the immune system following vaccination. Whereas canvaxgen covers genes responsible for expressing antigenic proteins or peptides within proteins, aiding in cancer vaccine development and analysis in cancer vaccine development and analysis (see Fig. 2).

Fig. 2

Ontological modeling of canvaxgens. The model illustrated the distinction between canvaxgen and antigen

As described in our recent CanVaxKB paper [20], we have identified 263 canvaxgens. Analyzing these genes using the Gene Ontology (GO) enrichment method revealed enriched features such as host-virus interactions of virus genes associated with cancer, cancer pathways, and tumor antigens [20]. This approach provides valuable insights that may not be accessible through a sole focus on peptide level.

Ontology development based on cancer vaccine ontology design pattern (ODP)

We added 704 cancer vaccine terms to the VO using the methods described in the Methods section. To ensure reproducible and consistent representation of cancer vaccine terms, we applied the cancer vaccine ODP to construct the cancer vaccine term ROBOT template in CSV format file (https://github.com/vaccineontology/VO/blob/master/src/templates/cancer_vaccine.csv). The ROBOT tool [27] was then employed to generate cancer vaccine terms in an OWL file. Reused OBO Foundry ontology terms such as disease, host/pathogen organisms were extracted from the source ontologies such as DOID, MONDO, and NCBITaxon using OntoFox tool [25]. These OWL files were merged in VO.

Entities associated with cancer vaccines shown in the ODP were implemented as logic axioms of a cancer vaccine in VO. Figure 3 provides an example of a specific vaccine ‘12MP vaccine’ implemented based on the ODP. Specifically, this vaccine:

Fig. 3

Protégé screenshot of a specific cancer vaccine in VO. The figure shows a specific cancer vaccine ‘12MP vaccine’ in VO that was generated based on ODP. It includes the term label, definition, definition source, status of vaccine development status in addition to logic axioms representing targeted organism and cancer type, vaccine platform, and vaccine administration route

‘is a’ ‘melanoma vaccine’

‘immunizes against disease’ some melanoma

‘immunizes organism’ some ‘Homo sapiens’

‘has role’ some ‘subunit vaccine role’

‘vaccinated in route’ some ‘intramuscular route’

In addition to the representing relations shown in the ODP, we added annotation properties, such as ‘definition source’ to include vaccine reference(s) and the CanVaxKB website link associated with a specific vaccine.

Cancer vaccines constitute a small portion of vaccines in VO. To facilitate working with a cancer vaccine only ontology, we generated the cancer vaccine view (cvo.owl) by retrieving all cancer vaccines and associated axioms from the full version VO. The cancer vaccine view is available on: http://purl.obolibrary.org/obo/vo/cvo.owl. As of March 12, 2024, the cvo.owl contains 928 classes with 681 specific cancer vaccines and 23 cancer vaccine terms such as ‘lung cancer vaccine’, ‘melanoma vaccine’ for term organization. It includes 744 VO specific classes and reused 154 classes from external ontologies (15 BFO classes, 92 DOID classes, 3 GO classes, 9 IAO classes, 2 MONDO classes, 26 NCBITaxon Classes, 15 OBI classes and 3 PATO classes). This CVO view also has 16,003 axioms including 3,706 logical axioms.

Additionally, a Makefile was generated to provide instructions to automate the above ontology development workflows, including ‘modules’ to convert template to OWL file, ‘imports’ to retrieve terms from external source ontologies and clean the axioms, ‘all’ to build a merged and inferred version of VO and generate OBO Foundry principle checking report, and ‘views’ to generate cancer vaccine view.

Analysis of cancer vaccines based on cancer disease types using DL query

350 cancer vaccines target cancer occurring in specific anatomical entities, with some vaccines targeting multiple types of cancer occurring in different anatomical sites. For example, ‘RAS Peptide Cancer Vaccine’ targets RAS peptide-specific anti-tumoral T-cell cytotoxic immune response, inhibiting cancer in colon, pancreas, and lung [48]. Figure 4 illustrates a DL query example that was used to retrieve all cancer vaccines targeting ‘reproductive organ cancer’ based on ‘immunizes against disease’ logic axioms.

Fig. 4

Protégé screenshot of a DL query in VO. The DL query is used to retrieve all the cancer vaccines that immunize against reproductive system cancer. The query is shown at the top right of the screenshot, while the results are shown at the bottom right

These 350 cancer vaccines target 11 types of ‘organ system cancers’ defined in Disease Ontology (DOID), including 104 vaccines against reproductive organ cancers, 61 vaccines against gastrointestinal system cancers, and 60 vaccines against hematologic cancers, etc. (Fig. 5).

Fig. 5

Hierarchical classification of cancers and cancer vaccines for specific cancers. The cancer hierarchy was generated using OntoFox tool and visualized using Protégé, and the numbers next to the cancer types were obtained using DL query for VO

According to DOID, cancer is classified not only by anatomical entities (i.e. ‘organ system cancer’) but also by the type of cell from which it is derived (i.e. ‘cell type cancer’). DOID contains 4 categories under the ‘cell type cancer’. Based on cell types, CVO has 236 cancer vaccines, including 210 melanoma vaccines, 22 carcinoma vaccines, 14 sarcoma vaccines, and 2 malignant astrocytoma vaccines (Fig. 5). Melanoma is a skin cancer that starts in the melanocytes [49]. Carcinoma is a malignant neoplasm of epithelial cell origin of the internal or external lining of the body [50]. Astrocytoma is a cancer of astrocytes that support and connect nerve cells in the brain and spinal cord [51]. A sarcoma is a malignant tumor that arises from cells of mesenchymal origin (e.g., bone, muscle, fibrous tissue) [52].

Figure 5 summarizes all cancer vaccines classified based on the cancer disease hierarchy. The number of the vaccines obtained using DL query based on ‘immunizes against disease’ targeted on different cancer types.

Analysis of cancer vaccines based on vaccine platform using SPARQL query

Cancer vaccines can be prepared by different vaccine platforms including subunit vaccine, DNA vaccine, recombinant vector vaccine, and live attenuated vaccine. These platforms are general vaccine development methods not unique for cancer vaccines. The vaccine platforms were defined using equivalent axioms that allowed the vaccines to be classified based on vaccine platforms in addition to other vaccine features such as immunized disease, targeted pathogen, etc. For example, ‘DNA vaccine’ is defined as:

def: A vaccine that is composed of a plasmid vaccine vector (a circular double stranded DNA molecule) containing the whole of parts of genes encoding one or more vaccine antigen proteins.

equivalent to:

‘vaccine by platform type’ and ‘has role’ some ‘DNA vaccine role’

We developed SPARQL query scripts to retrieve cancer vaccines of interest including those cancer vaccines based on specific vaccine platforms. For example, the following query is used to obtain all DNA cancer vaccines. The SPARQL query can be used to retrieve cancer vaccines of interest. For example, the following query is used to obtain all DNA cancer vaccines:

PREFIX cancer_vaccine:<http://purl.obolibrary.org/obo/VO_0000177>

PREFIX has_role: <http://purl.obolibrary.org/obo/RO_0000087>

PREFIX DNA_vaccine_role: <http://purl.obolibrary.org/obo/VO_0000623>

SELECT distinct? vaccine_label? vaccine.

FROM < http://purl.obolibrary.org/obo/merged/VO>

WHERE.

Running on the Ontobee SPARQL endpoint, the above script returned 86 DNA cancer vaccines with vaccine name and IRI, such as ‘Cancer DNA vaccine p.DOM-AH1 encoding fragment C’ (http://purl.obolibrary.org/obo/VO_0004425), ‘Melanoma DNA vaccine pN4a-MAGE-1-GM’ (http://purl.obolibrary.org/obo/VO_0004435), and ‘Lung metastasis DNA vaccine pCEP4-MUC1 encoding MUC1’ (http://purl.obolibrary.org/obo/VO_0004437). These DNA vaccines contain DNAs that encode different proteins or protein fragments and target different cancer types. While relational database-supported web queries may provide a more user-friendly interface, using SPARQL queries allows us to efficiently query ontology-stored knowledge using the standard query language from an ontology triple store.

Ontological modeling cancer vaccine clinical trial studies and applying to melanoma vaccine clinical trial data analysis

Cancer vaccines aim to stimulate a robust host response for effective immunotherapy and immunoprevention. The main goal of the host response is to achieve a sustained and high level of immune response through cancer vaccine, which includes immunity to specific antigenic epitopes [53]. About 70% of vaccines collected in CanVaxKB are in the development stage of clinical trials. However, it can be a challenge in evaluating the host responses to cancer vaccines, especially in clinical trials involving human subjects.

Response Evaluation Criteria in Solid Tumors (RECIST) is a standardized criteria by multiparty validation to assess changes in solid tumors. It is also commonly used as a means of evaluation in cancer immunotherapy and cancer vaccine outcomes [54, 55]. RECIST has adopted twelve criteria to quantify tumor growth and defines terms like complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD) to further classify tumor clinical trial response [56]. The RECIST criteria could also be used to study the associations with tumor long-term outcomes [54]. Therefore, we have adopted RECIST for cancer vaccine standardization outcome measurements.

OBO Foundry ontologies were used to represent main components and their relations in cancer vaccine clinical trials including assessment of cancer vaccine efficacy according to RECIST criteria. A cancer vaccine clinical trial involves participants, who are human study subjects with cancer and receive cancer vaccines as part of the trial. Each clinical trial study has a clinical trial identifier commonly used to reference the specific clinical trial on clinicaltrials.gov. The clinical trial generates outcome measurement data that are assessed according to RECIST criteria to draw conclusions regarding the efficacy of the cancer vaccine against the cancer (Fig. 6).

Fig. 6

Ontological modeling of cancer vaccine clinical trials. The model represented main processes in a clinical trial study and showed how cancer vaccine activity and efficacy were evaluated using RECIST criteria. The terms with prefix indicate the source of the ontologies. All ontologies used in the model are OBO Foundry ontologies. CTO: Clinical Trial Ontology, DOID: human Disease Ontology, IAO: Information Artifact Ontology, OBI: Ontology for Biomedical Investigations, and VO: Vaccine Ontology

The ontology model was applied to collect patient outcome measurements after melanoma vaccine treatment from the clinical.gov. Then the vaccine efficacies were analyzed using RECIST criteria. The RECIST endpoints were assigned through manually evaluated patient outcomes of 83 melanoma vaccine clinical trial studies. Figure 7 shows the frequency of the endpoint measurements. The results of the assessment illustrate the effect and activity of cancer vaccines. For example, a total of 6 cases of Anti-tumor response and 13 cases of immune response were reported among the 83 annotated melanoma clinical trial studies. This result reflects that the accessed melanoma vaccines work effectively on immune reactions to identify or attack abnormal cancerous cells and stimulate a cytotoxic T-cell reaction. Also, the immune system’s ability to generate a strong response significantly influences the likelihood of achieving a favorable outcome defined by RECIST.

Fig. 7

Occurrence of different patient outcome endpoints measurements of melanoma vaccines. The results generated from clinical trial endpoints analysis of 53 annotated melanoma cancer vaccine data. The endpoints were measured and categorized based on RECIST. The numbers located next to the bars indicated the frequency occurred for corresponding endpoint

The use case demonstrated that ontology models can guide the cancer vaccine clinical trial data collection for vaccine efficacy analysis. Standardization of the data and representation details of RECIST criteria using ontology may potentially enable the automated analysis of vaccine immune response in the future.