MarkerDB 2.0: a comprehensive molecular biomarker database for 2025

Abstract

MarkerDB (https://markerdb.ca) has become a leading resource for comprehensive information on molecular biomarkers. Over the past 3 years, the database has evolved significantly, reflecting the dynamic landscape of biomarker research and increasing demands from its user community. This year’s update, which is called MarkerDB 2.0, introduces key improvements to enhance the database’s usability, consistency and the range of biomarkers covered. These improvements include (i) the addition of thousands of new biomarkers and associated health conditions, (ii) the inclusion of many new biomarker types and categories, (iii) upgraded searches and data filtering functionalities, (iv) new features for exploring and understanding biomarker panels and (v) significantly expanded and improved descriptions. These upgrades, along with numerous minor improvements in content, interface, layout and overall website performance, have greatly enhanced MarkerDB’s usability and capacity to facilitate biomarker interpretation across various research domains. MarkerDB remains committed to providing a free, publicly accessible platform for consolidated information on a wide range of molecular (protein, genetic, chromosomal and chemical/small molecule) biomarkers, covering diagnostic, prognostic, risk, monitoring, safety and response-related biomarkers. We are confident that these upgrades and updates will improve MarkerDB’s user friendliness, increase its utility and greatly expand its potential applications to many other areas of clinical medicine and biomedical research.

Graphical Abstract

Graphical Abstract.

Introduction

The term biomarker was first introduced in the 1970s (1). Over that time, the identification and validation of biomarkers for various research and clinical decision-making applications has gained increasing attention. Indeed, the importance of biomarkers to the fields of medicine, medical research and drug testing is clearly reflected in the enormous number of publications containing the term ‘biomarker’ or ‘biomarkers’ found in PubMed – which now numbers >1.2 million (2). A biomarker is broadly defined as a ‘characteristic’ that is measured as an indicator of normal biological processes, pathogenic processes or responses to an exposure or intervention (3). These characteristics may be molecular, histological, radiographic or physiological. Biomarkers may be single, well defined, quantitatively measurable entities, such as blood glucose, or a collection of hard-to-define, qualitative characteristics, such as histological or radiographic image features that require specialized interpretation and training. Biomarkers can be molecular (genes, proteins and metabolites), microscopic (karyotypes and histological images) or macroscopic (visible phenotypic features or readings that the naked eye can see). Despite their many and varied characteristics, there is now a general consensus on how biomarkers may be grouped or categorized (3). Specifically, according to the Biomarkers, EndpointS and other Tools (BEST) framework (3), there are seven major classes of biomarkers: (i) diagnostic, (ii) monitoring, (iii) pharmacodynamic/response, (iv) prognostic, (v) safety, (vi) susceptibility and (vii) risk.

Given the enormous variation in the type, character and quality of biomarkers, it is easy to understand why detailed information about clinically useful biomarkers is often hard to find and even harder to understand. Combined with the huge proliferation of biomarker studies over the past two decades, accessing comprehensive, up-to-date and useful biomarker data, especially for molecular biomarkers, remains challenging. Indeed, it is not uncommon to see the same biomarkers being repeatedly ‘rediscovered’, completely contradictory biomarker results being presented or new biomarkers being sought out when outstanding biomarkers already exist. This is especially true with molecular biomarkers where the advent of ‘omics’ measurements has led to thousands of new molecular biomarkers appearing in clinical tests and many more investigational biomarkers appearing in the literature.

It is because of the confusing state of the literature on biomarkers and the enormity of the number of known or suspected molecular biomarkers that MarkerDB was created. First released in 2021, MarkerDB was designed to simplify the challenge of navigating biomarker space by offering free, consolidated and user-friendly information on a wide range of molecular biomarkers in an easily searchable and readily downloadable format. MarkerDB’s definition of a molecular biomarker is relatively broad and includes a diverse array of biological molecules covering chromosomes, DNA, RNA, proteins, lipids and small molecules (both exogenous and endogenous). MarkerDB also focused on providing detailed data on a subset of clinically approved molecular biomarkers for which there was extensive information available, namely clinical diagnostic, risk, monitoring, safety, response and prognostic biomarkers.

As one of the first, comprehensive, consolidated resources on molecular biomarkers to be made freely available, MarkerDB has been well received, having already been cited ∼100 times. However, as with any initial database release, MarkerDB was also a work in progress. Over the past 3 years, our team has received invaluable feedback about MarkerDB’s strengths and weaknesses from its many users. We have also carefully reviewed MarkerDB’s content and identified a number of oversights and shortcomings with version 1.0 of the database. For instance, several well-known clinical markers were missing or incompletely described in version 1.0. Likewise, thousands of genetic risk [single nucleotide polymorphism (SNP)] biomarkers were not included or formally presented in an easily viewable format. Similarly, the number of previously defined exposure biomarkers was far too limited and certainly needed updating. Likewise, the depth and detail of many disease descriptions as well as several biomarker descriptions needed improvement and more extensive referencing. Additionally, MarkerDB’s layout and data formatting was outdated, clumsy to search/navigate and inconsistent compared with our other databases such as HMDB (4) and MiMeDB (5).

In addition to this internal evaluation, we also did a detailed survey of other recently published biomarker databases to further refine MarkerDB’s position in the biomarker ‘universe’. Specifically, we looked at databases such as CellMarker (6), Lnc2Cancer (7), Cancer Genome Interpreter (8), Exposome-Explorer (9), ResMarkerDB (10) and TheMarker (11). This was done to assess the size, strengths, weaknesses, graphical user interface design, layout and biomarker focus areas of these databases and to compare them with MarkerDB.

These internal and external assessments allowed us to design a new and improved version of MarkerDB—MarkerDB 2.0. To create this new version of MarkerDB the curation team primarily focused on making significant additions and improvements to the database’s content. These included adding 12 674 new risk biomarkers, 45 new safety biomarkers, 636 new monitoring biomarkers and 25 new response biomarkers. In total, 694 clinically approved as well as 6329 investigational biomarkers were added to the database. In addition to these data enhancements, the old MarkerDB website design was completely revamped and modernized. This involved the creation of a new and improved layout offering refined searching, filtering and display features. This redesign will allow users to navigate through the database and find or interpret biomarker information much faster and far more easily. Other improvements to MarkerDB 2.0 include more complete, fully referenced biomarker and condition descriptions, an updated, more consistent Disease Ontology (DO) and the addition of many more external and internal hyperlinks (and references). Data on biomarker panels (i.e. collections of >1 biomarker) have also been enhanced to provide comprehensive information on individual biomarkers within each panel, displayed in a tabular format with clickable links. Collectively, these changes have further distinguished MarkerDB 2.0 from most other biomarker databases. Indeed MarkerDB 2.0 is now quite unique in terms of its size and scope, particularly through its inclusion of both clinically approved and investigational biomarkers across multiple diseases, conditions and molecular marker types as well as across all major biomarker categories (see Table 1).

Table 1.

Comparison of MarkerDB 2.0 with other biomarker databases currently available

Database	Biomarker categories covered	Focus areas	Marker molecules	Special features
MarkerDB 2.0	Six out of seven BEST categories	Clinically approved and investigational biomarkers	Proteins, DNA, RNA, karyotypes, small molecules and other chemicals	Provides concentration values, AUC curves and ROC data for many biomarkers. Advanced search options, 3D structure viewers and detailed biomarker descriptions.
CellMarker	Cell-specific markers and features	Cell type markers in human and mouse tissues	Proteins, genes and other cell-specific molecules	Includes web tools for analyzing single-cell sequencing data.
Lnc2Cancer	Long non-coding RNAs (lncRNAs) in cancer	Cancer-related lncRNAs and circular RNAs (circRNAs) markers	lncRNAs and circRNAs	Allows users to filter data by cancer subtype, lncRNA/circRNA type and clinical relevance.
Cancer Genome Interpreter	Cancer-related biomarkers	Cancer mutation markers and therapeutic options	Cancer genes and cancer drugs	Utilizes OncodriveMUT, a rule-based approach for assessing the tumorigenic potential of mutations.
TheMarker	Five out of seven BEST categories	Therapeutic biomarkers	Proteins, DNA, miRNAs and lncRNAs	Artificial intelligence-enhanced data that support applications in drug development and clinical practice.
Exposome-Explorer	Environmental and diet biomarkers	Diet, environment and lifestyle exposure biomarkers	Small molecules and other chemicals	Extensive referencing and quantitative data. Aids in understanding how external environmental factors influence disease risk and progression.
ResMarkerDB	Response biomarkers to cancer monoclonal antibody treatment	Biomarkers for response to cancer treatment	DNA, noncoding DNA and RNA	Provides information on biomarkers associated with the response to cancer therapies.

More detailed descriptions regarding each of these improvements to MarkerDB are given under the following subsections: (i) more biomarker entries, (ii) improved layout, design and usability (iii) biomarker panel, search and filtering improvements and (iv) database implementation, curation and FAIRness.

More biomarker entries

Since its release in 2021 (12), the MarkerDB 2.0 curation team has dedicated thousands of hours to biomarker research, collection and data uploads. More than 6700 biomarkers, 300 health conditions and over 4000 concentration values have been added since the previous version was published (Table 2). Uploading these new biomarkers and new concentration values required extensive literature research, meticulous formatting and thorough vetting by the curation team. Only peer-reviewed papers where biomarkers met strict selection criteria—namely that the biomarkers had to be quantitative and molecular. This rigorous selection process significantly reduced the number of relevant papers from 1.2 million to 10 738.

Table 2.

Summary of biomarkers, biomarker classes, molecular categories and word counts that have changed substantially from MarkerDB 1.0 to MarkerDB 2.0

Biomarker data	MarkerDB 1.0	MarkerDB 2.0	Number increase	Percent increase (%)
Total number of biomarkers	27 686	34 483	6510	24.6
Total number of genes	930	3212	2282	245.4
Total number of proteins	142	218	76	53.5
Total number of chemicals	1090	1664	574	52.7
Genetic biomarkers	26 300	32 447	6147	23.4
miRNA biomarkers	0	95	95	new
ct/cfDNA biomarkers	0	101	101	new
Health conditions	699	992	293	41.9
Diet-related biomarkers	21	117	96	457.1
Risk biomarkers	0	12 674	12 674	new
Safety biomarkers	0	45	45	new
Monitoring biomarkers	0	636	636	new
Response biomarkers	0	25	25	new
Reference concentrations	2385	5932	3547	148.7
Clinical biomarker concentrations and measurements	37 573	41 522	3949	10.5
Clinically approved biomarkers	26 434	27 128	694	2.6
Investigational biomarker concentrations and measurements	4179	11 357	7178	171.8
Investigational biomarkers	1550	7879	6329	408.3
Average word count per description	153.4	214.5	61.1	39.8

To identify relevant studies for investigational biomarkers, the following medical subject headings terms were used in various combinations: ‘Biomarkers’, ‘Metabolites’, ‘Metabolomics’, ‘Genetic Markers’, ‘Proteomics’, ‘Genomics’, ‘Metabolic Networks and Pathways’, ‘Bioinformatics’, ‘Molecular Diagnostic Techniques’, ‘Metabolic Diseases’, ‘Randomized Controlled Trials’, ‘Chemical Compounds’, ‘Data Integration’ and ‘Enzymes’. Boolean operators (AND, OR) combined terms, such as ‘Biomarkers’ AND ‘Metabolic Diseases’ or ‘Genetic Markers’ AND ‘Metabolomics’. The search covered studies published between 1994 and 2024 without language restrictions and included studies identifying or validating investigational biomarkers in human health, RCTs, cohort studies and those providing quantitative biomarker data, such as genome-wide association studies (GWAS) or bioinformatics integration. Non-peer-reviewed, animal-only and in vitro studies lacking human relevance were excluded. The literature review process was streamlined using several in-house software tools to semi-automatically sort through the biomarker papers and pull data for manual reviewing, cleaning, formatting and uploading.

Chemical and protein biomarkers

A significant focus for MarkerDB 2.0 was on expanding the number and type of exposure biomarkers, especially diet-related biomarkers and pollutant (i.e. chemical exposure) biomarkers. Pollutant and other chemical exposure biomarkers were sourced primarily from the National Health and Nutrition Examination Survey (NHANES) (13), PubMed (2) and our own publications on exposure/diet biomarkers (14). NHANES (13) provided detailed information and biofluid concentration values for hundreds of exposure chemicals in the US population, accounting for a substantial portion of the new chemical exposure biomarkers. Diet-related, or ‘food biomarkers’ were sourced primarily from the literature, with a focus on published ‘biomarkers of food intake’ (15) as well as the excellent data in NHANES (13). In total, over 300 exposure biomarkers were added to MarkerDB 2.0 and were recategorized into safety, monitoring or response biomarkers, and hundreds of existing ones were updated with new descriptions, concentration values and statistical information. For version 2.0, exposure biomarkers have been recategorized into response, safety and monitoring to align with the BEST biomarker framework (3).

Another major focus of MarkerDB 2.0 was on expanding the number of clinically approved biomarkers. A particularly rich source of clinically approved diagnostic biomarkers was the Mayo Clinic biomarker database (https://www.mayoclinic.org/diseases-conditions). Data from the Mayo Clinic provided 125 new, clinically approved biomarkers including 73 new chemical biomarkers and 52 new protein biomarkers. Other clinically approved biomarker data were also retrieved from the Catalog of Somatic Mutations in Cancer – COSMIC (16) providing 101 new biomarkers including 76 new clinical diagnostic biomarkers. In addition to this, clinically approved biomarker data were sourced from websites of testing service providers such as Mobile Health (https://www.mobilehealthconsumer.com), Foundation Medicine (https://www.foundationmedicine.com), Guardant Health (https://guardanthealth.com), ThermoFisher Oncomine (https://www.oncomine.com/solutions) and Exact Sciences (https://www.exactsciences.com). Significant effort was put into properly annotating, manually cross-checking and thoroughly updating the information on the biomarkers collected from these websites.

Genetic biomarkers

Another major focus for MarkerDB 2.0 was the inclusion of additional genetic biomarkers, particularly SNP ‘risk’ biomarkers generated via GWAS. While very few GWAS markers have made it into the clinic, they are of general interest for disease risk assessment and have been used in many consumer-oriented genetic testing services. Given their potential importance, we collected a significant number of GWAS markers or polygenic marker panels and added them to MarkerDB 2.0. Using data from our previously published GWAS-ROCS (17) database, over 3000 new genetic or polygenic risk biomarkers for >170 conditions were collected, updated and added to MarkerDB 2.0. Data retrieval from GWAS-ROCS was streamlined using an in-house tool specially developed by the MarkerDB team, enabling automatic data extraction via web scraping and database access. These multi-SNP genetic risk biomarkers are technically considered biomarker panels and are treated as such when displayed and formatted in MarkerDB 2.0. Newer GWAS biomarkers and biomarker panels, not found in GWAS-ROCS, were sourced from the NHGRI-EBI Catalog of human genome-wide association studies (18). These newer GWAS datasets were downloaded, reformatted and processed using the GWIZ package (https://bioinformaticshome.com/db/tool/GWIZ) to generate appropriate biomarker panels, statistical performance information and receiver operating characteristic (ROC) curves. This work led to the cumulative addition of 6397 SNP genetic risk markers across 266 medical conditions.

In addition to these SNP biomarkers, we also revisited our genetic mutation biomarker data. While MarkerDB’s collection on mutation biomarkers is already very extensive, we were able to collect new data on clinically significant cancer mutation biomarkers from My Cancer Genome (19). These mutation biomarkers were identified as clinically significant based on data from clinical trials and drug responses. A total of 24 new genetic biomarkers for seven cancer conditions were added to MarkerDB 2.0. SNPedia (20), dbSNP (21) and various papers in PubMed were used to gather and write additional details on all the new SNPs as well as the newly added mutation biomarkers, along with information on their corresponding genes or intergenic regions.

MarkerDB 2.0 has also added RNA, circulating tumor DNA (ctDNA) and cell-free DNA (cfDNA) biomarkers, which were not present in version 1.0. The addition of these molecular biomarkers contributed 196 new entries, including 71 clinical and 125 investigational biomarkers, covering a variety of conditions. The COSMIC database (16) also provided ctDNA and cfDNA biomarkers for genes confirmed to be involved in cancer. Additionally, a number of microRNA (miRNA) biomarkers were sourced from PubMed (22).

Counting both clinically approved and investigational biomarkers, a total of 574 new chemical biomarkers and 76 new protein biomarkers were added to MarkerDB 2.0. This led to the inclusion of 325 new medical conditions and the updating of 362 existing medical conditions in MarkerDB 2.0. Additional performance information [sensitivity, specificity, areas under the curve (AUCs), etc.] about many of these new protein and chemical biomarkers was generated via MetaboAnalyst (23). MetaboAnalyst, through its ‘Biomarker Analysis’ module was used to calculate relevant statistics for several biomarkers, especially when key data were missing from the original source material. Specifically, ROC curves, suitable combinations of biomarkers (i.e. biomarker panels) and related biomarker performance statistics were generated for many of these new protein and chemical biomarkers where possible.

Normal reference values

While normal reference values (or reference intervals) for proteins or metabolites in various biofluids are not typically considered biomarkers, they are vital to many diagnostic protocols and for interpreting diagnostic tests. Adult reference intervals for many important chemical biomarkers or metabolites have long been available in reference books and journals and have been compiled in the HMDB (4) for nearly two decades. However, information on pediatric reference values is much harder to find. This is because establishing pediatric reference intervals is challenging due to difficulties in recruiting large pediatric populations and the need for multiple age partitions to account for growth and development (24). Major changes in organ maturation, body weight, body length, physiology and immune functions as children grow mean that their reference intervals for many proteins and metabolites can undergo major changes too. Indeed, pediatric reference intervals can be profoundly different from adult reference intervals for the same biomarker (24). Therefore, relying solely on adult reference values for children can be problematic, occasionally leading to dangerous misdiagnoses (24).

The CALIPER project (24) was established ∼20 years ago to address this issue by creating a database with clinical reference ranges carefully measured for 14 000 healthy children, spanning a wide range of ages (from 0 to 18 years) and covering various protein and metabolite biomarkers in different biofluids. This database defines reference concentration intervals for plasma, serum and blood, as the 2.5th and 97.5th percentiles in a healthy reference population, considering factors like age, sex, body mass index, ethnicity, time of collection and analytical methods (24). The CALIPER effort has contributed hundreds of important baselines or normal concentration values in plasma, serum and blood for a total of 92 metabolites and 65 proteins. All of these molecules are considered as clinically important for accurate pediatric health or diagnostic assessments.

The CALIPER pediatric reference intervals have been integrated into MarkerDB 2.0, adding 905 new biomarker concentration values. These reference interval additions should help support more precise interpretation of pediatric biomarker data, reducing the unnecessary follow-up tests and improving clinical decision making. By including these values, MarkerDB reaffirms its commitment to providing comprehensive, age-appropriate reference data wherever possible. Indeed, the CALIPER dataset was recently complemented by our own clinical reference interval data for urine collected on a population of healthy neonates (25). In addition to these reference values for infants and children, another 1000 healthy adult (male and female) reference interval values for metabolites in serum, plasma, urine and cerebrospinal fluid were compiled from the HMDB 5.0 (4) and added to MarkerDB 2.0. In all cases, detailed descriptions of the biomarkers are available by clicking on the MarkerDB ID to visit the individual biomarker pages.

To allow these values to be more easily accessed by users, a ‘Browse Reference Concentrations’ page has been added, providing detailed information on healthy concentrations or normal states for all biomarkers in MarkerDB 2.0. Users can access this page from the MarkerDB 2.0 landing page and can view concentrations either directly on the page, or by clicking on the MarkerDB ID.

Improved biomarker and health condition descriptions

Significant improvements have been made to the quality of both biomarker and health condition descriptions in MarkerDB 2.0. Every entry in MarkerDB 2.0 now includes a detailed description, typically ranging from 100 to 500 words, providing comprehensive information about each biomarker and its related health condition or conditions along with numerous, in-line references or citations. In total, the curation team has enhanced descriptions for over 300 biomarkers and over 600 conditions using both semi-automated and manual annotation approaches.

For well-known health conditions and biomarkers, ChatGPT 4.0 (https://openai.com/index/gpt-4) was used to generate initial descriptions, which were then manually reviewed and edited by two separate curators who added appropriate references and manually improved the style, accuracy and readability. The ChatGPT prompts used for generating these initial descriptions were carefully designed to ensure scientifically accurate outputs, with a focus on key elements such as the name, structure, properties, function, uses and indications for each biomarker. A typical prompt was: ‘Using scientifically appropriate language covering aspects pertaining to structure, function, uses and indications, explain in a single 300 to 400-word paragraph what X is.’

Approximately 80% of the ChatGPT-generated outputs were usable with revisions to ensure accuracy and scientific rigor, while 20% did not meet our acceptance criteria and were manually rewritten. For less familiar conditions and biomarkers, descriptions were manually researched using a variety of well-regarded reference sources (26–31) and then manually written. As before, the manually generated descriptions were reviewed and edited by at least two separate curators. This rigorous curation process ensured that all descriptions were well-written, comprehensive, informative and reliable. Additionally, hundreds of hyperlinked references have been integrated directly into the descriptions, allowing users to access original sources and conduct further reading without leaving the website.

Improved layout, design and usability

Improved layout

The MarkerDB 2.0 website has undergone a comprehensive redesign to enhance its visual appearance, improve its layout, simplify its navigation, upgrade its usability and ameliorate the presentation of useful biomarker data. This updated version (https://markerdb.ca) stands apart from its previous iteration (https://old.markerdb.ca) in a number of areas. Notably, the new navigation bar (located at the top of each page) now features four key tabs: ‘Browse’, ‘Search’, ‘Downloads’ and ‘About’. Users can now search, browse and explore all conditions and all biomarkers directly from the landing page without needing to select a specific biomarker category – a feature that was not previously available. A screenshot of this new landing page is provided in Figure 1A. By clicking on the corresponding biomarker category images or using the ‘Browse’ dropdown menu, users can explore 13 different biomarker options, including ‘All Biomarkers’, ‘Condition Specific Biomarkers’, ‘Chemical Biomarkers’, ‘Genetic Biomarkers’, ‘Protein Biomarkers’, ‘Karyotype Biomarkers’, ‘Diagnostic Biomarkers’, ‘Prognostic Biomarkers’, ‘Risk Biomarkers’, ‘Safety Biomarkers’, ‘Monitoring Biomarkers’, ‘Response Biomarkers’ and ‘Reference Concentrations’. The main biomarker types (condition specific, chemical, genetic, protein and karyotype) are further categorized as diagnostic, prognostic, risk, safety, monitoring or response (Figure 1A). The updated layout further enhances clarity and accessibility, enabling users to interpret results with greater ease.

Figure 1.

Screenshots from MarkerDB 2.0 showcasing (A) the MarkerDB 2.0 landing page, (B) the MarkerDB 2.0 chemical biomarkers table and (C) concentrations and statistics for the chemical biomarker d-Glucose. Not all fields are shown.

In the previous version of MarkerDB (12), risk biomarkers were categorized as predictive biomarkers. This nomenclature was at odds with a recent ‘BEST’ consensus document on biomarker categorization (3) which defines a risk biomarker as a type of biomarker that indicates the likelihood of developing a disease or medical condition in an individual who is currently asymptomatic. This recategorization brings MarkerDB 2.0 into better alignment with modern, consensus biomarker definitions. It should be noted that ‘Exposure’ is not a biomarker category included in this BEST consensus document. Therefore, the exposure category has been recategorized into either monitoring, safety or response.

Detailed information about MarkerDB 2.0’s collection of biomarkers is now provided in sortable tables, which list biomarker names, MarkerDB IDs (with hyperlinks), associated health conditions and other relevant features (Figure 1B). Users can click on column headers to sort the selected biomarker data alphabetically or numerically. To further explore individual biomarkers, users can click the ‘MarkerDB ID’ button to view comprehensive information and statistics about any specific biomarker, including related medical conditions(s), references, links to other database like FooDB (https://www.foodb.ca) and HMDB (4) along with other details about the biomarker itself. One of the new biomarker features that MarkerDB 2.0 now offers includes interactive 3D structural images for all protein and small molecule biomarkers. These are viewable via the Mol*Viewer (32).

Improved design and enhanced user experience

MarkerDB 2.0’s redesign significantly improves the user’s ability to retrieve and work with data, focusing on enhancing clarity, navigation and data accessibility. The new, consistent color scheme and standardized iconography simplify navigation, allowing users to intuitively explore biomarker categories and health conditions. These improvements, including higher resolution and color-coordinated icons, guide users effortlessly through the database, reducing the time spent searching for specific information. The new icons, representing each of the 11 different biomarker types and categories, are far more intuitive in design and are of higher quality than those in the previous version (Figure 1A). Similar standardized iconography was also used to generate icon images corresponding to the 22 broad health conditions covered in MarkerDB 2.0. These health conditions are described in more detail later in this document.

Each biomarker entry is now more richly annotated with descriptions, structural or schematized images, hyperlinks, references and most importantly, quantitative data. This includes P-values, concentration cutoff values, sensitivity/specificity values, AUC values and ROC curves (Figure 1C).

Interactive structure viewer

Recent developments in 3D protein structure prediction via tools like AlphaFold (33), along with continued advances in experimental structure determination [and the corresponding growth in the Protein Data Bank (PDB) (34)] allowed the MarkerDB 2.0 curation team to obtain high-quality coordinate data for all 218 protein and peptide markers in MarkerDB 2.0. In assembling this structure database, priority was given to experimentally determined structures from the PDB (34), while lower priority was given to AlphaFold-predicted structures housed in the AlphaFold Protein Structure Database (APSD) (33). If no structure was available in either the PDB (34) or the APSD (33), the MarkerDB 2.0 curator team ran individual AlphaFold 2.0 predictions for the known biomarker sequence on a local GPU computer with a local, fully validated version of AlphaFold 2.0. Therefore, in addition to the full sequence data being available for the peptide and protein biomarkers, 3D coordinate data (in the form of PDB formatted files) and 3D images are also available for both viewing and downloading. Specifically, MarkerDB 2.0 supports interactive 3D viewing of colored ribbon diagrams of all peptide and protein biomarkers via the Mol*Viewer (32), a high performance open-source molecular visualization tool (Figure 2). With the Mol*Viewer, the structure quality and structure annotations are displayed in both functional and biological contexts. The Mol*Viewer is also used by PDB (34) and PDBe (35).

Figure 2.

Screenshot showing a 3D protein structure for glycated hemoglobin from the Mol*Viewer. This feature is accessible by clicking ‘View 3D Structure’ when on a given protein biomarker page.

In addition to displaying 3D protein structures, the 3D structure of all small molecule biomarkers in MarkerDB 2.0 is now interactively viewable via the same Mol*Viewer (32). Coordinate data in the form of SDF and MOL files are also available for download. This interactive structure viewing capability is new to MarkerDB 2.0 and was not available in version 1.0 (12).

Biomarker panels, search and filtering improvements

Biomarker panels

A biomarker panel comprises a group of two or more markers that collectively reflect various pathophysiological processes associated with a disease or health condition. In the initial release of MarkerDB, biomarker panels were displayed similarly to individual biomarkers, but also associated with a logistic equation composed of multiple normalized biomarker levels instead of conventional concentration values. Recognizing the importance and frequency with which biomarker panels are being described, especially with GWAS data, we prioritized enhancing their usability and searchability in MarkerDB 2.0. This was done by reformatting the layout of the ‘old’ biomarker panels so that they would be more easily searched and displayed in a user-friendly tabular format. Each biomarker panel in MarkerDB 2.0 now includes direct links to individual biomarker pages, allowing rapid access to detailed information about each biomarker in the panel. An example of the new biomarker panel format can be seen in Figure 3. In addition to these enhancements, we also added over 450 new biomarker panels to MarkerDB 2.0.

Figure 3.

Screenshot from MarkerDB 2.0 of an example of a chemical biomarker panel for d-Glucose. Not all fields are shown.

Biomarker panels in MarkerDB 2.0 are identified by querying metabolites specifically linked to the condition of interest. These panels may either come from pre-existing publications or be generated by our team. To calculate logistic tables and AUC values, we use the means and standard deviations of biomarker concentrations to construct synthetic populations, which are then split into training and testing sets. A logistic regression model is trained on the training set, and ROC curves are generated. Sensitivity, specificity and AUC values are derived from the test set, with variability accounted for by adjusting the mean of each ROC curve point by two standard deviations. It is worth noting that individual biomarkers within a panel are not double-counted in categories.

Database search improvements

Advanced filtering options and refined search capabilities allow users to quickly locate relevant biomarkers by class, approval status, name, sequence or structure, streamlining data retrieval and boosting efficiency. For instance, an additional filter for health conditions has been added to the ‘Browse Conditions’ page, which displays schematic icons representing each health condition (Figure 4A). This new filter allows multiple health conditions to be selected simultaneously. On MarkerDB 2.0’s biomarker category pages, users can refine their searches using checkbox-style filters based on chromosome number, biomarker type, biofluid, organism, protein function type, clinical versus investigational biomarkers, etc. These filters vary slightly across different biomarker types or categories and are present at the top of each biomarker page. Additionally, specific chromosome filters are now available for genetic and karyotype biomarkers (Figure 4B).

Figure 4.

Screenshots from MarkerDB 2.0 demonstrating: (A) new health condition icons and filters from the ‘Condition Biomarkers’ and ‘Browse Conditions’ pages, and (B) filter options shown on the ‘Genetic Biomarkers’ page. Filters vary in their content and layout depending on the biomarker category. Not all fields are shown.

Health condition ontology

A biomarker does not necessarily have to be associated with a disease. Indeed, biomarkers can also be associated with different states of health, such as pregnancy, menstrual cycle phases or menopause. As a result, we have chosen to use a more general health status term, ‘health condition’ or simply ‘condition’, to refer to health states associated with biomarkers. MarkerDB 2.0 has reorganized its collection of health conditions for better clarity and improved categorization. Previously, in MarkerDB 1.0, health conditions were linked to broader categories listed under ‘General Conditions’. Version 2.0 introduces refinements to these categories based on insights and standards described in the DO, developed and maintained by disease-ontology.org (36).

Specifically, the curation staff with MarkerDB 2.0 has corrected the names of a number of existing health condition categories, removed less informative ones and added new health categories to match standard DO categories. The broad health condition/disease category ‘Other’ has been eliminated, with all health conditions reclassified into more specific and more appropriate categories. A ‘Browse Conditions’ page now displays these new health condition categories (Figure 4A) and provides access to all biomarker data relevant to each health condition which is also depicted with an appropriate, easy-to-understand icon. Along with new icon images, associated health conditions are displayed in biomarker pages following a tree-like layout showing each condition subcategories. When hovered over, these icons show the condition name for quick identification or disambiguation. Clicking on a specific health condition icon and selecting ‘Apply Filter’ will filter the results, displaying the health sub-categories along with their corresponding biomarkers.

Database implementation, curation and FAIRness

MarkerDB 2.0 was developed using the Ruby on Rails web framework (https://rubyonrails.org, version 2.5.1). It employs a MariaDB relational database (https://www.mariadb.com, version 10.4.6) to manage biomarker and health condition data. This includes entity relationships, external references, comprehensive descriptions, chemical and protein structures and concentration values. The design follows the Model-View-Controller architecture, ensuring modularity and efficient separation of data logic from user input and presentation (37). MarkerDB 2.0 is converted into web pages via an HTML interface responder, providing users with an intuitive and user-friendly experience. The dedicated server hosting MarkerDB 2.0 runs the Ubuntu 22.04 OS on a computer equipped with a 2.3 GHz CPU including >50 GB of disk space and 2 GB of RAM offering scalability and smooth performance. Weekly backups and stringent data management protocols safeguard data integrity.

All data in MarkerDB 2.0 has been vetted and validated by a team of experts, each with at least an undergraduate degree in relevant fields such as genetics, molecular biology, biochemistry or natural products chemistry. Data entry is carefully monitored through a centralized, password-controlled database, allowing any changes and edits to the database to be consistent, time-stamped and automatically transferred. Curators received extensive training by the lead curator(s) in biomarker annotation via hands-on mentoring, text instructions, peer support and tutorials.

MarkerDB 2.0 adheres to FAIR principles (Findable, Accessible, Interoperable and Reusable) (38), with compliance details provided under the ‘About’ menu tab. Each biomarker has a unique and permanent MDB identifier for easy findability, and the data are accessible via web pages and the download pages. Interoperability is ensured by making the data available in standard TSV and XML format. The database is well documented and curated to ensure reusability.

MarkerDB 2.0 offers a robust web-based user-interface with extensive search functionalities and an application programming interface (API) for programmatic data access. The database undergoes continuous improvements and updates, with minor corrections and additions being done on a regular basis. The data in MarkerDB are released under a Creative Commons Attribution BY and NC license.

Conclusion and future directions

We believe MarkerDB 2.0 represents a significant advancement for online biomarker databases. This version of MarkerDB is characterized by a number of important biomarker additions, enhanced search and display functionalities, improved data organization, expanded accessibility and improved biomarker panels. These biomarker additions include numerous clinical biomarker reference intervals (for adults and children) and the addition of over 6000 new biomarkers, primarily chemical and protein biomarkers, from extensive literature searches and the consolidation of data from a number of online sources such as NHANES (13), the Mayo Clinic (https://www.mayoclinic.org/diseases-conditions), CALIPER (24) and genetic biomarkers from GWAS-ROCS (17). The comprehensive redesign and the implementation of several advanced database search and display features should make MarkerDB 2.0 a much more user-friendly and informative resource than its predecessor.

While this particular update to MarkerDB has emphasized the addition of new categories and types of biomarkers, the expansion of current biomarker categories, and improvements in layout to aid in usability, future updates of MarkerDB will likely focus on expanding the content of other biomarker categories such as prognostic biomarkers and further expanding the newly added response, monitoring and safety biomarkers. Currently, MarkerDB features only molecular biomarkers. Future updates are expected to expand into other biomarker types, including imaging and tissue/histology biomarkers to provide a more holistic view of both clinical biomarkers and research biomarkers. These additions will, no doubt, bring new challenges concerning biomarker data compilation and display. Regardless of where our future plans take us, the MarkerDB team remains committed to implementing continuous improvements, addressing user feedback and enhancing MarkerDB’s functionality, reliability and usability. By staying responsive to community needs, we aim to ensure that MarkerDB remains an invaluable resource for medical researchers and healthcare professionals.

Data availability

All the data from MarkerDB 2.0 are freely accessible through the MarkerDB 2.0 website at https://markerdb.ca or the dedicated MarkerDB 2.0 download page. The platform is open and free to use, with data downloads compatible with all modern web browsers. Download files are available in TSV and XML formats. Additionally, chemical biomarkers can be accessed in multiple formats, including SMILES, SDF, MOL, PDB, InChI and InChIKey. Protein structure data are available in PDB format while sequence (gene and protein) data are stored in FASTA format. Images are available in PNG format, and all textual data and metadata are written in English. The data in MarkerDB are released under a Creative Commons Attribution BY and NC license.

Funding

Genome Alberta (a division of Genome Canada); Canada Foundation for Innovation (CFI); Natural Sciences and Engineering Research Council of Canada (NSERC). Funding for open access charge: Genome Alberta.

Conflict of interest statement. None declared.