Human disease biomarker panels through systems biology

Abstract

As more uses for biomarkers are sought after for an increasing number of disease targets, single-target biomarkers are slowly giving way for biomarker panels. These panels incorporate various sources of biomolecular and clinical data to guarantee a higher robustness and power of separation for a clinical test. Multifactorial diseases such as psychiatric disorders show great potential for clinical use, assisting medical professionals during the analysis of risk and predisposition, disease diagnosis and prognosis, and treatment applicability and efficacy. More specific tests are also being developed to assist in ruling out, distinguishing between, and confirming suspicions of multifactorial diseases, as well as to predict which therapy option may be the best option for a given patient’s biochemical profile. As more complex datasets are entering the field, involving multi-omic approaches, systems biology has stepped in to facilitate the discovery and validation steps during biomarker panel generation. Filtering biomolecules and clinical data, pre-validating and cross-validating potential biomarkers, generating final biomarker panels, and testing the robustness and applicability of those panels are all beginning to rely on machine learning and systems biology and research in this area will only benefit from advances in these approaches.

Introduction

In the ongoing quest for personalized medicine and ever more precise medical tests, molecules referred to as biomarkers have shown great potential. Life at a very fundamental level is a complex mixture of DNA, RNA, proteins, lipids, and metabolites. With the activation and suppression of biological processes, both beneficial and detrimental to an organism’s wellbeing, these biological molecules can also vary, not only in quantity, but also regarding their shape, interactions, location, and activity. More concisely, the World Health Organization has defined a biomarker to be “any substance, structure, or process that can be measured in the body or its products and influence or predict the incidence of outcome or disease” (Organization and Safety 2001). According to an early definition (Frank and Hargreaves 2003), for a biological molecule to be useful as a biomarker, it should fit into one of the three following types: type 0, which represents a patient’s natural biology with perturbations being directly correlated with symptoms, though this may include correlation and not necessarily causation; type 1, which represents the effects of treatment, being directly associated with a biological mechanism affected by the therapeutic molecule; and type 2, which represents a molecule that is directly and unambiguously associated with a patient’s feelings, biological functions, or chance of survival and which can be objectively used as a clinical data point. Type 2 biomarkers are also referred to as surrogate endpoints, as they step in for traditional clinical endpoints, in either a responsive or predictive sense. While these three types of biomarkers still encompass the field and the overall definition has not changed much, several subcategories with more precise definitions have evolved over the last two decades. A remapping of data points as predictive, diagnostic, prognostic, or therapeutic biomarkers has gained popularity (Fong and Winter 2012; Carlomagno et al. 2017; Ankeny et al. 2018; Verstockt et al. 2019; Chang and Ladame 2020; Nguyen et al. 2021; Tian et al. 2021), clearly categorizing them by their purpose and function.

Modern science has brought about many new biomarkers that can be used in research and in the clinic; however, various factors have slowed or even inhibited their entry into the health sector, and have been extensively discussed (Mayeux 2004; Agache and Rogozea 2017; Hampel et al. 2018; Locke et al. 2019). Cost is one such factor, compounded by several sources such as purchasing and maintaining the equipment used for identifying and quantifying biomolecules, various reagents, and hiring trained specialists in that specific technique. Also, along these lines is profitability, since funding agencies look at performance, patentability, market fit, and overall need and demand when identifying which biomarkers to support. Next, variations in populations and subpopulations, resulting both from geography and non-regional environmental and lifestyle factors, can also limit scalability; only biomarkers that stand up to rigorous testing conditions can progress to the clinic. Moreover, the source of a particular biomarker also has to be considered for clinical implementability. Non-invasive sources of biological material such as saliva, urine, and epithelial cells (i.e. cheek swab), as well as imaging such as fMRI, are easily obtained. Minimally invasive techniques like drawing blood are also commonly used for biomarkers, though other sources like internal organ tissue and cerebrospinal fluid are sometimes the only possibility. All sources of biological material possess distinct sets of molecules, each able to answer different biological questions. Genetic markers can easily be identified with non-invasive buccal swabbing and some metabolic biomarkers can be collected from urine or saliva, though a proteomic marker might need to be measured from blood or other tissues. Other classes of biomarkers, such as imaging data from MRI scans, though non-invasive, may be costly and time-consuming. Even when covering all these topics, regulatory agencies must see a clear benefit for the use and implementation.

Despite these drawbacks and limitations, biomarkers as a whole show great potential in the health sector by introducing new clinical tests and offering alternatives to methods that rely on phenotype and symptoms alone. Especially in psychiatry, patients often exhibit symptoms that overlap with multiple disorders and classification by symptoms alone can also create very heterogeneous groups of patients, all expected to receive the same treatment options (García-Gutiérrez et al. 2020). Of course, biomarker tests are not intended to replace medical professionals, but rather assist them by integrating genetics, behavioral science, proteomics, and other areas with the phenotypic data that they analyze when calculating risks, benefits, prognoses, and treatment options (Venigalla et al. 2017).

Biomarkers in human diseases

One of the primary compounding factors when trying to discover biomarkers for human diseases is the inherent variability between individuals and within the same individual over time due to genetic, environmental, and endogenous factors (Aylward et al. 2014; Enroth et al. 2014; Thyagarajan et al. 2016; Aziz et al. 2019). Differences in biological molecules can arise from ethnicity, sex, and lifestyle choices like smoking and alcohol use. As such, any biomarker must pass stringent testing and be replicable, reliable, and robust to ensure the lowest rate possible of false negatives and false positives. Several categories of data have entered the field to predict the risk of an individual to develop a disease, detect the disease at an earlier stage, identify or confirm a disease, and predict a patient’s response to a particular treatment and their prognosis.

The first biomarker tests were originally possible by studying a single biological molecule, and in their elegant simplicity, are still widely used to this day. In 2000, specific types of DNA mutations were linked to their environmental causes (Shugart 2000); a year later, mutations in certain protein-encoding DNA sequences increased the risk for an individual to develop cancer, and an array of genetic, epigenetic, and protein data points were suggested as potential biomarkers to identify cancer (Srinivas et al. 2001); and over the following few years, other biomarkers had been identified to measure the progression of Alzheimer’s disease in a drug trial (Jack et al. 2003), predict atherothrombotic events (Ridker et al. 2004), measure cellular senescence (Krishnamurthy et al. 2004), identify metabolic syndrome (Ryo et al. 2004), and assess oxidative stress (Dalle-Donne et al. 2006). Since these initial studies, the number of publications involving biomarkers has skyrocketed, having reached a total of over one million publications with the keyword biomarker in PubMed (NCBI Resource Coordinators 2018). The path taken to research biomarkers, however, shifted with the confirmation that many diseases and disorders cannot be boiled down to a single dysregulated gene, protein, or metabolite, which is especially true with psychiatric disorders, since a combination of environmental factors and genetic factors often leads to their development (Kubota et al. 2012; Meyer-Lindenberg and Tost 2012; Klengel and Binder 2015).

Each class of biomarker calls for a different technique for analysis, which can also vary depending on the properties of the biomarker itself. RNA in all its forms, proteins and polypeptides, lipids, and metabolites can all be analyzed differently (see Fig. 1). More relevant to the patient, each type of biomarker is well-suited to respond to certain physiological questions (see Table 1). Genomics can be used for predictive, prognostic, and diagnostic biomarkers, which may allow a medical professional to predict disease risk; however, it usually cannot be used for therapeutic biomarkers, since DNA rarely varies in response to treatment or medication, excluding gene therapy. In contrast, RNA and protein levels are dynamic and are associated with a vast array of biological processes, making them suitable for all four classes of biomarkers. Metabolites also show great potential as diagnostic, prognostic, and therapeutic biomarkers (Tolstikov et al. 2020). Lastly, non-molecular biomarkers, such as visual phenotyping via radiographic capture, circulating tumor cells (Punnoose et al. 2010; Cabel et al. 2017; Pantel et al. 2019; Yang et al. 2019), physiological parameters like blood pressure and resting heart rate, as well as the human microbiota (Manor et al. 2020; Julie et al. 2021), all have their niches in various aspects of biomarker use.

Fig. 1.

Possible sources for molecular biomarkers with each molecule type categorized depending on the method(s) by which they are most commonly detected. Created using open

source software at Diagrams.net. Different classes of biomarkers are separated based on the method(s) available for detection. Nucleic acid sequencing can be used for DNA, epigenetics by DNA modification, mRNA, post-transcriptional modifications, along with tRNA, snRNA, miRNA, siRNA, and aRNA. Immunoassays can detect epigenetics by histone/chromatin modification, polypeptides, proteins, and post-translational modifications. Mass spectrometry can detect all the biomarkers in immunoassays with the addition of metabolite and lipids. UV detection can also be used for metabolites. Gas chromatography and nuclear magnetic resonance can be used for lipids

Table 1.

Different classes of molecular biomarkers, subclasses, and some published applications

Molecule	Biomarker	Common applications
Nucleic acids	Genetic	-Risk factors and predisposition
	-DNA	-Diagnosis and prognosis
		-Therapy efficacy
		-Drug dosing (absorption and metabolism)
	Transcriptomic	-Diagnosis and prognosis
	-RNA	-Determining physiological states
		-Identifying infection states
	Epigenetic	-Diagnosis and prognosis
	-DNA	-Risk factors and predisposition
	-Histones	-Therapy
	-Chromatin
Amino acids
	Proteomic	-Diagnosis and prognosis
	-Proteins	-Therapy
	-Polypeptides
	-Post-translational modifications
Metabolites	Metabolomic	-Diagnosis and prognosis
	-Hormones	-Treatment response
	-Free amino acids	-Forensic toxicology
	-Drug metabolites	-Drug efficacy/toxicity
	-Vitamins
Lipids	Lipidomic	-Diagnosis and prognosis
	-Phospholipids	-Treatment response
	-Glycerides	-Risk factors and predisposition
	-Sterols
	-Fatty acids

During the ongoing quest for biomarkers, new data points are constantly being investigated, both in terms of identity and in origin. Despite a great deal of studies regarding protein levels in diseases, as well as the well-established use of protein-based biomarkers in the clinic, protein post-translational modifications (PTMs) have entered the clinic without much fanfare (García-Giménez et al. 2017). PTMs affect several aspects of protein function and properties, modifying localization, binding, activity, and stability, among other characteristics (Ramazi and Zahiri 2021). Each modification can be added and removed in different ways, is reversible or irreversible, enzymatically regulated or spontaneous, site-specific or global, and is controlled by various regulatory pathways (Walsh et al. 2005). PTM data is an untapped source of information for biomarkers and proteomic studies overall since PTMs can respond rapidly to stimuli, as PTMs can occur more quickly than protein translation, simultaneously offering a more phenotypic profile than transcriptomic data. The different mechanisms of addition and removal and their associated biological pathways make PTMs even more attractive for biomarker studies, with the enzymes themselves also being potential biomarkers for some diseases and disorders. There are hundreds of known protein modifications and, though some are more consistently found, such as sulfide bridges between cysteine residues, many can play roles both upstream and downstream to protein-based dysregulations in diseases.

Not all applications have been implemented in the clinic; this is not an exhaustive list of applications or references. Documented uses of biomarkers that are genetic (Novelli et al. 2008; Ziegler et al. 2012; Coppedè et al. 2014; Zhang et al. 2019; Center for Drug Evaluation and Research 2021), transcriptomic (Heidecker et al. 2011; Pedrotty et al. 2012; van Rensburg and Loxton 2015; Xi et al. 2017; Zhang et al. 2019), epigenetic (Kubota et al. 2012; Coppedè et al. 2014; Li et al. 2014; García-Giménez et al. 2017; Soler-Botija et al. 2019), proteomic (Hewitt et al. 2004; Theodorescu et al. 2006; Egerer et al. 2009; Humphries et al. 2014; Raemdonck et al. 2014; Thelin et al. 2017; Zhang et al. 2019; Zhao et al. 2020), metabolomic (Helander and Beck 2005; Chen et al. 2011; Zhang et al. 2015; Tam et al. 2017; Wang et al. 2020; Long et al. 2020), and lipidomic (Chandler et al. 2016; Kim et al. 2017; Yan et al. 2017, 2018; Aquino et al. 2018; El-Ansary et al. 2020; Liu et al. 2020).

Phosphorylation, for example, is a widely studied regulator of protein activity and has been studied in relation to multiple diseases including identifying tumors (Carter et al. 2020) and neurofibrillary tangles in Alzheimer’s disease (Buerger et al. 2006; Henriques et al. 2016). Other targets include multiple PTMs potentially related to Parkinson’s disease (Schmid et al. 2013) and the complex field of glycoproteins, which has shown promise in identifying neoplastic diseases (Díaz-Fernández et al. 2018). PTMs are also known to play roles in aging and age-related diseases (Santos and Lindner 2017) and inflammatory processes (Yang et al. 2017), and dysregulated modification of tubulin has been suggested to be behind a wide range of human diseases including cancer, cardiac diseases, and bleeding disorders (Magiera et al. 2018). As not all modifications are performed enzymatically, cellular conditions can also affect how proteins are modified; succinylation, malonylation, formylation, succination, and acetylation are all hypothesized—or have been proven—to be possible without any transferase (Lin et al. 2012). Documenting these changes has thus far led to insight into the metabolic (de)regulation and metastasis that occurs in gastric cancer (Song et al. 2017) and GAPDH malonylation has been suggested to play a role in inflammation in macrophages (Galván-Peña et al. 2019). Overall, PTMs show great potential as markers of phenotypes while still maintaining flexibility and dynamicity due to their fast and often reversible nature.

Unfortunately, due to variations between individuals (Enroth et al. 2014) and within individuals over time (Cicognola et al. 2015), the infeasibility of measuring biomarkers directly in more sensitive tissue like the brain or other internal organs, and the fact that some more complex, multifactorial diseases like psychiatric disorders stem from a vast number of small changes (2009; Genovese et al. 2016; Selzam et al. 2018), not every potential biomarker can exhibit such stark differences as those seen in the some of the first clinical biomarkers. In response, the scientific community has found ways to use biomarkers with lower robustness. In one approach, a biomarker test can be prescribed a complementary technique in parallel with other analyses to distinguish between similar diseases, confirm suspicions, or objectively measure disease progression or treatment efficacy. Another approach is biomarker panels, a set of independent biomarkers that, together, provide a more robust and reproducible answer to a given medical question. As computer processing capacity increases, so does the complexity of problems that can be solved, and systems biology has now made it feasible to sift through tens of thousands of data points to return a set of potential biomarkers that strengthen one another. Overall, biomarker panels are able to perform a slightly different function from individual biomarkers, since the panel is additionally able to help exclude compounding factors and work with multifactorial disorders with more certainty.

In one example, a panel of metabolites has been used to concisely map a profile for the healthy metabolic response to physical activity (Netzer et al. 2011). In another study, over 1600 plasma and urine protein levels were studied for associations with dozens of human diseases (Dudley and Butte 2009). Such studies have not been limited to being reactive, but have also been proposed to be applicable for proactive, long-term health monitoring and other facets of personalized medicine (Miller et al. 2019). In conjunction with magnetic resonance imaging, transcriptomic data of 30 different genes were used to predict a progression from clinically isolated syndrome to multiple sclerosis (Tossberg et al. 2013). Trends in fasting plasma glucose levels can be predicted with a panel of 9 metabolites, despite neither the individual metabolites nor a panel of standard risk factors such as BMI and age being able to accurately model fasting plasma glucose levels (Hische et al. 2012). Another panel of metabolites can predict dysregulation in fasting plasma glucose levels, type-2 diabetes, and insulin resistance (Menni et al. 2013), sometimes even a decade or more before developing a disorder (Tabák et al. 2009; Rhee et al. 2011; Wang et al. 2011). A three-part panel for detecting colorectal cancer and pre-malignant colorectal neoplasia has also been approved using genetic, epigenetic, and protein assays (Imperiale et al. 2014a, b). Using NMR-based metabolomic data and multivariate statistics, Song et al. compiled 82 references to form a list of potential and identified biomarkers in over 10 disease categories before classifying the biomarkers for categories like diagnosis, treatment, and prognosis (Song et al. 2019) and Dhama et al. compiled a comprehensive list of several types of potential biomarkers for physiological and psychological stress, along with their sources and possible clinical significance (Dhama et al. 2019). Volatile organic compounds are also being investigated using gas chromatography mass spectrometry, with the potential to identify cancers, genetic and metabolic disorders, infectious diseases, and various other conditions (Buljubasic and Buchbauer 2015). Lastly, comprehensive tissue-based DNA sequencing panels have been approved by the FDA as companion diagnostic tools to identify cancer types and suggest compatible therapy options (U.S. Food & Drug Administration 2021) and new panels are being tested and approved every year.

Among the many classes of human diseases, psychiatric disorders are among those that most need biomarkers, since the clinically observable symptoms and genetic background for many disorders are often similar and have significant overlap (Doherty and Owen 2014). This makes an early, fast, and precise diagnosis through clinical observation alone extremely difficult. Moreover, antipsychotics, which are used to alleviate symptoms in conditions such as schizophrenia and bipolar disorder, do not have a universally positive response. While one patient might respond well to a given medication, another patient may not respond, or may even have a negative response (Leucht et al. 2013). Repeated negative responses to treatment paired with debilitating side effects increase patient stress and can lead to treatment dropout (Wahlbeck et al. 2001; Rabinowitz and Davidov 2008; Rabinowitz et al. 2009), reducing the patient’s quality of life and incurring more medical costs. Another important point is that, for many psychiatric disorders, by the time symptoms appear and can be effectively diagnosed and treated, the patient’s quality of life may have already been negatively, potentially irreversibly, affected (Guest et al. 2016). Biomarker panels for diagnosing neurological and psychological conditions in the prodromal phase or earlier, distinguishing between two or more similar psychiatric disorders, and predicting a patient’s response to a given treatment option without needing to perform a trial-and-error testing method would all be extremely beneficial to patients and healthcare providers alike.

Generating biomarker panels

Unlike the less fruitful searches for molecular signatures related to psychiatric disorders, diagnostic biomarkers for other diseases, such as cancer, have been established with relative success (Parkes et al. 1995; Cramer et al. 2011; Fontecha et al. 2016). Potential diagnostic biomarkers for ovarian cancer, for example, were discovered after the analysis of five predictive models across three combined stages of validation and discovery, compared against an already well-established biomarker for some types of cancers, the CA-125 protein (Zhu et al. 2011). Although this is a reliable method for establishing new biomarkers, in the absence of any previously established biomarkers, mathematical modeling has paved new avenues in biomarker development.

Establishing biomarkers is a multistep and multidisciplinary process that can be split into three major stages before passing through regulatory bodies into the clinic: discovery, development and validation, and implementation (Fig. 2) (Martins-de-Souza et al. 2011; Menetski et al. 2019). During the discovery and development/validation stages, the selection criteria for patient eligibility for a predictive model form a key step for efficiently achieving statistical significance. Thus, establishing biomarkers requires not only an understanding of the biochemical pathways involved in the disease or biological state by analyzing molecular data but also properly collecting clinical data from patients for selection and stratification, as well as for establishing exclusion and validation criteria (Laifenfeld et al. 2012; Drucker and Krapfenbauer 2013; Fröhlich et al. 2018; Golubnitschaja et al. 2018). Overall, the discovery step seeks to do just that: collect potentially vast amounts of data from various biomolecular and clinical assays to analyze and filter for potentially relevant data points. Molecular data from different omics and clinical information are used to compose molecular signatures after increasingly advanced statistical processing steps involving advanced computational methods, algorithms, and machine learning. When using single omics methods, molecular signature investigations benefit from multivariate approaches with penalties for classification of features or regression. However, in the case of higher complexity multi-omics, the lasso or elastic net selection methods help to efficiently reduce data dimensionality (Bravo-Merodio et al. 2019; Freue et al. 2019). Moreover, algorithms such as tree ensembles can combine different models to identify potential molecular signatures using distinct techniques (Chen et al. 2013; Toth et al. 2019; Shuwen et al. 2020; Acharjee et al. 2020). Mathematical models are being successfully applied to a small portion of omic datasets, and without them, the search for biomarkers would take a prohibitive amount of time or even find dead ends. Despite great advances, however, there are still no computational methods or data interpretation protocols that are so well established that they have become state-of-the-art for molecular signatures.

Fig. 2.

A workflow describing the steps going from a need for a biomarker-based test to a qualified biomarker test for clinical use, passing through an unbiased discovery stage, a biased development and validation stage, and an implementation stage before being proposed for acceptance and clinical use. Created using the open

source software at Diagrams.net. A flow chart detailing the process from a need for a biomarker-based test, through genomic, epigenomic transcriptomic, proteomic, metabolomic, lipidomic, PTMomic, and clinical data in a discovery-based, unbiased approach into biomolecular and clinical data. Using univariate and multivariate statistics, this and the previously mentioned data can both be published in publicly available repositories and peer-reviewed articles. Statistically filtered data can then pass into potential biomarkers that loop through more statistics and AI-based filtering using machine and deep learning before passing through a model adjustment, cross-validation, and reproducibility tests to reach biomarker status. This then passes on to a biomarker-based test, all of which occurred in a validation-based, biased approach. During implementation, the assay may be redeveloped and is eventually tested for clinical implementability and reaches clinical test status. In the acceptance and use phase, approval by regulatory agencies must be obtained to reach a final, qualified biomarker test for clinical use

After the discovery phase, the development and validation phase is often composed of two verification rounds to further narrow down the data points highlighted during the first phase. The first, cross-validation round reduces the number of potentially false biomarkers through sets of algorithm retraining, which may be based on similar samples to those used in the discovery phase or based on a new set of samples in an omics or multi-omics experiment (Smit et al. 2007; Singh et al. 2019; Chierici et al. 2020). During this round, one common method of pre-validation is to randomize and divide the retraining samples into a number k of groups of similar sizes, leading to its name k-fold cross-validation. In this method, one group is set aside as a validation set, while the computer models a fit to the training set. During cross-validation, the number of potential targets in the biomarker panel is reduced as the model is adjusted, contributing to increased specificity and sensitivity thresholds, along with a greater power of discrimination between biological states (Harris et al. 2009).

Next, a second validation round is where the greatest challenge lies when establishing biomarkers due to both the greater number of samples required for analysis and a dependence on long-term resources to carry out studies over a potentially great period of time (Bonassi et al. 2001). During this stage, many new samples must be collected to analyze the potential biomarkers. A biomarker or panel can undergo different levels of validation to be applicable for more restricted or broader populations, as well as conditions under which it is valid. Furthermore, biomarker validation must also include confirmatory clinical endpoint analyses to prevent misleading conclusions (Strimbu and Tavel 2010). During development, statistical methods can be used to predict a model's error when applied to new cases, such as in the double cross-validation method (Smit et al. 2007; Szász et al. 2016). After this phase, the rigor of the methods used to establish the potential biomarkers and the accessibility of the tool for clinical use are considered. To ensure smooth translatability to the clinic, a great deal of care must be taken at every step of the discovery process, especially when creating sample groups, working with missing values, selecting assays, and testing for robustness and scalability across subpopulations (Mnatsakanyan et al. 2018). Meeting the criteria for a particular regulatory and implementing body, such as the FDA in the United States or the EMA in Europe, is a complex and time-consuming procedure as well, as these clinical trials must carefully weigh the risks and benefits of the use of the biomarker panel. Robustness must be proven, and time must be taken to determine the precision of the test; false negatives and false positives must be at a minimum or the panel may be sent back for statistical refinement, requiring a new round of clinical trials and once the panel of biomarkers has passed these trials, it may then be qualified for eventual implementation (Manolis et al. 2015; Menetski et al. 2019). To ensure a robust test, future biomarker investigations should be carried out with multi-omics techniques on several layers of biological systems, also providing a better understanding of the molecular mechanisms involved and adding extra dimensions to the mathematical model being developed. While this makes the process of building a panel of biomarkers more challenging, it creates a more useful test with a higher power of separation.

Conclusions

The early days of biomarker research brought important advances for a few select cases, dominated by individual biomolecules. As different clinical targets received focus, the field of biomarker research quickly shifted once single-molecule biomarkers began to slow and new target classes were scrutinized through transcriptomics, lipidomics, PTMomics, and metabolomics. New targets for disease risk and predisposition, diagnosis, prognosis, and treatment are constantly being discovered before passing through rigorous testing steps to ensure reproducibility and population-wide applicability. Though the era of individual biomarkers has not come to a close, biomarker panels have entered the scene to increase the robustness of tests for clinical implementation. Biomarker panels bring together various data points, even using multi-omic approaches, to compose tests to complement current clinical practices. By using the tests in a controlled manner, medical professionals can efficiently diagnose, rule out, confirm, or distinguish between various possible diseases, allowing a patient to be more quickly and accurately diagnosed and therefore treated quickly and effectively. Multifactorial diseases, such as psychiatric disorders, with a culmination of both genetic and environmental factors have already benefited greatly from such biomarker panels and show great potential for further development. Tests to facilitate the diagnosis and treatment of such disorders are of great importance to doctors and patients alike, improving the quality of life of the patient and simultaneously reducing the cost of medical care.

Propelling this new chapter of biomarker research is advanced computational approaches, able to process and filter data in ways that are simply not feasible to be performed manually or with simple algorithms. Machine learning has found a niche in biomarker research, performing vital steps during both the discovery and validation phases of panel generation. Pre-validation, selection methods, and cross-validation of data all contribute to a high-quality panel of potential biomarkers with higher certainty of achieving implementable results and a lower chance of having confounding factors influencing the results. As computational methods and processing speed advance, increasingly large datasets involving multiple data sources and collection methods will be accessible. Though this field is still in its infancy with no widely standardized methods for data processing or validation, it is nonetheless paving the way for new biomarker-based tests with various clinical applications.

Funding

This research is funded by the Coordination for the Improvement of Higher Education Personnel (CAPES; grant number 88887.495565/2020–00) and The São Paulo Research Foundation (FAPESP; grant numbers 2016/07948–8, 2017/25588–1, 2018/03422–7, 2019/25957–2, and 2020/04746–0).

Declarations

Conflict of interest

The authors declare no competing interests.