Multi-Omic Early Cancer Diagnosis: Integrating Proteins, Metabolites, and Tumor DNA into a Unified Detection Framework

1. Introduction: From Single-Analyte Biomarkers to Systems-Level Diagnosis

The clinical trajectory of a malignant neoplasm is determined less by its histological label than by the stage at which it is discovered. Localized disease is frequently curable through surgical resection or definitive radiotherapy, whereas metastatic dissemination reduces five-year survival for most solid tumors to single or low double-digit percentages. This stark disparity has made early detection one of the highest-yield objectives in modern oncology. Yet the classical screening modalities — mammography, low-dose computed tomography, colonoscopy, and organ-specific serum markers — remain limited by anatomical scope, procedural burden, imperfect sensitivity, and, in several settings, an inability to demonstrate a mortality benefit.

The emergence of multi-omic liquid biopsy offers a fundamentally different strategy. Rather than interrogating a single anatomical site or a single molecular class, multi-omics seeks to reconstruct the biological state of a tumor by simultaneously measuring complementary information layers circulating in peripheral blood: somatic and epigenetic alterations in cell-free DNA, tumor-derived and host-response proteins, and low-molecular-weight metabolites reflecting cellular biochemistry.

The premise is not that any one of these layers is sufficient. It is that their information content is largely non-redundant, and that their joint analysis can overcome the sensitivity-specificity trade-off that has constrained every single-analyte screening test to date. This article examines the biological foundations, analytical platforms, computational architecture, clinical evidence, and translational limitations of multi-omic early cancer detection.

2. Biological Foundations: What a Tumor Actually Releases into the Circulation

A tumor is not an isolated mass of proliferating cells. It is a dynamic organ system that exchanges material with the host through shedding, secretion, necrosis, apoptosis, and active vesicular transport. Each of these processes leaves a distinct molecular trace.

2.1. Cell-Free and Circulating Tumor DNA

Cell-free DNA (cfDNA) is a heterogeneous population of fragmented genomic DNA present in plasma, predominantly derived from the apoptosis of hematopoietic cells. In a patient with cancer, a small fraction — circulating tumor DNA (ctDNA) — originates from malignant cells.

Three properties of cfDNA are diagnostically exploitable:

• Sequence alterations. Somatic single-nucleotide variants, insertions/deletions, and fusion breakpoints provide high-specificity tumor signatures.
• Epigenetic modifications. DNA methylation patterns are cell-type-restricted. Malignant transformation disrupts these patterns, and because methylation retains a memory of cellular origin, it enables not only cancer detection but also tissue-of-origin (TOO) inference.
• Fragmentation architecture. cfDNA is not randomly cleaved. It is released in nucleosome-protected fragments of approximately 160-180 base pairs, with additional sub-nucleosomal species. Tumor-derived fragments are frequently shorter and exhibit altered end motifs, reflecting the aberrant chromatin accessibility and nuclease activity of malignant cells.

The central quantitative problem is fractional abundance. In early-stage disease, ctDNA may constitute less than 0.01% of total cfDNA. Detecting a variant present at one allele in ten thousand therefore requires error suppression strategies far beyond the capabilities of conventional sequencing.

2.2. The Plasma Proteome

Proteins enter the circulation through three routes: active secretion by tumor cells, passive release following cell death, and the host’s systemic response to malignancy. The third category is conceptually important. A small tumor may release negligible quantities of a tumor-specific antigen, yet provoke a substantial and measurable alteration in inflammatory, coagulation, complement, and acute-phase proteins.

This distinction underlies a strategic shift in biomarker discovery: from searching exclusively for tumor-derived signals to measuring the host’s reaction to the tumor, which may be amplified, earlier, and more reproducible.

2.3. The Metabolome

Metabolic reprogramming is now recognized as a core enabling characteristic of cancer. The Warburg effect — aerobic glycolysis — coexists with glutamine addiction, increased one-carbon flux, altered lipid desaturation, and rewired nucleotide synthesis. These changes generate detectable perturbations in circulating metabolite pools.

Certain mutations produce highly specific oncometabolites. Mutant IDH1 and IDH2 acquire a neomorphic activity that converts α-ketoglutarate to D-2-hydroxyglutarate (D-2HG), a molecule detectable in plasma and in tumor tissue by specialized spectroscopic methods. Similarly, loss of succinate dehydrogenase or fumarate hydratase leads to the accumulation of succinate and fumarate, respectively.

2.4. Complementary Information Layers

3. The Genomic and Epigenomic Layer: Reading the Tumor’s Informational Signature

3.1. Error-Suppressed Sequencing

Modern ctDNA assays rely on unique molecular identifiers (UMIs). Each original DNA molecule is tagged with a random barcode before amplification; all PCR products derived from that molecule are subsequently collapsed into a single consensus sequence. This procedure eliminates the majority of polymerase- and sequencing-induced artifacts.

In duplex sequencing, both strands of the original double-stranded molecule are independently barcoded and sequenced. Only variants present in both strand families are retained, reducing the effective error floor to approximately 10⁻⁴-10⁻⁵ mutant allele frequency. Cancer Personalized Profiling by deep Sequencing (CAPP-Seq), developed at Stanford University, demonstrated that personalized panels targeting the known mutational field of a patient’s tumor could detect ctDNA with exceptional sensitivity.

3.2. Clonal Hematopoiesis: The Dominant Biological Confounder

A critical and often underestimated source of false-positive results is clonal hematopoiesis of indeterminate potential (CHIP). With advancing age, hematopoietic stem cells acquire somatic mutations — particularly in DNMT3A, TET2, ASXL1, and TP53 — and expand clonally. The resulting mutant DNA is released into plasma and is indistinguishable from tumor DNA at the sequence level.

The accepted mitigation is matched white blood cell sequencing: variants present in the leukocyte compartment are annotated as hematopoietic in origin and excluded from the cancer signal. Failure to perform this correction is a recognized methodological defect in many early ctDNA studies.

3.3. Methylation as a Tissue Barcode

Unlike mutations, which are largely stochastic, DNA methylation is a regulated, cell-type-specific program. Cancer alters these programs in characteristic ways — hypermethylation of tumor-suppressor promoters combined with global hypomethylation.

Targeted methylation sequencing allows interrogation of thousands of differentially methylated regions while retaining enough genomic context to infer the tissue of origin. This capability is decisive for clinical utility: a screening test that reports «cancer detected» without indicating where to look generates an unmanageable diagnostic cascade.

3.4. Fragmentomics: Information Encoded in Where DNA Breaks

The seminal work of the Chinese University of Hong Kong group demonstrated that plasma DNA fragment lengths are not uniform across disease states; in hepatocellular carcinoma, tumor-derived fragments are systematically shorter than background cfDNA.

Subsequent research at Johns Hopkins University formalized this observation into DELFI (Digital Extraction of the Liquid Fragmentome Interface), a framework that uses genome-wide fragmentation patterns as a classifier. Because fragmentation reflects nucleosome positioning — and therefore transcriptional activity — it captures a functional epigenetic state without requiring bisulfite conversion or immunoprecipitation.

4. The Proteomic Layer: Quantifying Phenotype and Host Response

4.1. The Dynamic Range Problem

Human plasma contains proteins spanning more than ten orders of magnitude in concentration. Albumin and immunoglobulins alone account for roughly half of the total protein mass, while candidate tumor markers may exist at picogram-per-milliliter levels. This compressive dynamic range has historically limited proteomic discovery.

Three technological approaches have partially resolved this constraint:

1. Aptamer-based proteomics (SomaScan). Modified DNA aptamers bind proteins with high affinity, enabling multiplexed quantification of over a thousand targets.
2. Proximity extension assays (Olink PEA). Paired antibodies bearing DNA oligonucleotides generate a unique DNA barcode only upon simultaneous binding, permitting highly specific, low-volume multiplexing.
3. Data-independent acquisition mass spectrometry (DIA-MS). Unbiased fragmentation of all precursor ions within defined windows enables retrospective extraction of thousands of peptide signals.

Large-scale plasma proteomic studies in the UK Biobank have now established the genetic architecture and phenotypic associations of circulating proteins at population scale, providing an invaluable reference framework for biomarker interpretation.

4.2. The CancerSEEK Proof of Principle

The CancerSEEK assay, developed at Johns Hopkins University, remains the canonical demonstration of protein-DNA integration. It combined a targeted sequencing panel covering recurrent mutations in sixteen genes with a panel of seven circulating proteins (including CEA, CA19-9, CA125, AFP, hCG, HER2, and SCC).

In a retrospective case-control cohort comprising patients with nonmetastatic disease across eight cancer types, the combined algorithm achieved a median sensitivity of approximately 70% at a median specificity of approximately 98%. Critically, the protein and DNA components contributed non-redundant information: some tumors shed predominantly mutant DNA, others predominantly protein, and the union of the two substantially outperformed either alone.

This result established the central thesis of multi-omic detection. It did not, however, establish screening efficacy, since case-control designs do not reproduce the prevalence, spectrum, and workflow of a true screening population.

4.3. Why Classic Tumor Markers Underperform

Markers such as CA125, CEA, and CA19-9 suffer from two independent failure modes. First, they are not cancer-specific: benign inflammatory, infectious, and endocrine conditions elevate them. Second, they are not cancer-sensitive: many malignancies never express them at detectable levels. Multi-omic proteomics addresses both problems by replacing a single marker with a multivariate pattern and by incorporating host-response biology that is not contingent on tumor secretion.

5. The Metabolomic Layer: Interpreting the Functional State of the Tumor

5.1. Analytical Platforms

Metabolomics employs liquid chromatography-tandem mass spectrometry (LC-MS/MS), gas chromatography-mass spectrometry (GC-MS), and nuclear magnetic resonance (NMR) spectroscopy. Each offers distinct trade-offs: LC-MS provides broad coverage and high sensitivity; GC-MS delivers superior reproducibility for volatile and derivatized small molecules; NMR offers exceptional quantitative robustness and structural information at the cost of sensitivity.

5.2. Biological Signals of Malignancy

Metabolic reprogramming generates several classes of detectable perturbation:

• Glycolytic flux. Increased glucose consumption and lactate production alter circulating sugar and organic acid pools.
• Glutaminolysis. Enhanced glutamine utilization depletes circulating glutamine and enriches downstream tricarboxylic acid cycle intermediates.
• Lipid remodeling. Altered fatty acid desaturation and phospholipid turnover produce characteristic lipidomic signatures.
• Oncometabolite accumulation. IDH, SDH, and FH mutations generate structurally distinctive molecules with high mechanistic specificity.

Unlike genomic alterations, which are relatively static within a clone, metabolite concentrations reflect the instantaneous physiological state of the tumor and its microenvironment. This makes metabolomics uniquely suited to capturing functional heterogeneity.

5.3. The Confounding Problem

Metabolomics is simultaneously the most informative and the most confounded omic layer. Plasma metabolite concentrations are influenced by:

• Recent food and alcohol intake;
• Circadian phase and time of draw;
• Renal and hepatic function;
• Physical activity;
• Gut microbial co-metabolism;
• Medications and supplements;
• Fasting status and body composition.

A metabolite signature that discriminates cancer cases from controls in a single center may therefore encode nothing more than differences in fasting protocol or antibiotic exposure. Rigorous metabolomic studies must therefore employ standardized pre-analytical collection, matched dietary and medication documentation, and external validation across independent sites.

At present, no metabolite-only panel has achieved the level of prospective screening validation demonstrated by methylation-based or combined protein-DNA assays. Metabolomics is most plausibly positioned as a complementary layer within a multimodal model, rather than as a standalone screening instrument.

6. Multi-Omic Integration: Statistical and Computational Architecture

6.1. Levels of Integration

Multi-omic integration is not a single methodological operation. It occurs at three distinct levels:

Data-level (early) integration. Raw or normalized features from all layers are concatenated into a single feature matrix and analyzed jointly. This maximizes the opportunity to detect cross-layer interactions but suffers from dimensionality, heterogeneous scaling, and missing-modality problems.

Feature-level (intermediate) integration. Each modality is reduced to a lower-dimensional latent representation or a modality-specific risk score. These representations are then fused. This approach balances information retention with statistical efficiency and is currently the most common practical architecture.

Model-level (late) integration. Separate classifiers are trained for each modality, and their outputs are combined through voting, stacking, or a meta-learner. This is the most strong to modality-specific noise and missing data, but may discard valuable cross-layer interaction signals.

6.2. Machine Learning in Multi-Omic Detection

Algorithms deployed in this setting range from regularized logistic regression and elastic net to gradient-boosted decision trees, random forests, and deep neural networks. More recent architectures employ multimodal transformers that learn attention-based weights across omic layers.

Three methodological failures are recurrent in the literature and must be actively prevented:

1. Feature selection leakage. If biomarker selection is performed on the entire dataset before cross-validation, the reported performance is optimistically biased. Selection must occur inside each training fold.
2. Center and batch confounding. Omic platforms exhibit systematic batch effects. Without explicit batch correction and independent site validation, a classifier may learn the identity of the sequencing facility rather than the presence of cancer.
3. Calibration failure. A model may discriminate well (high area under the ROC curve) while producing poorly calibrated probabilities. In screening, where decisions depend on absolute risk thresholds, calibration is as important as discrimination.

6.3. The Bayesian Constraint: Why Specificity Dominates

The most important quantitative principle in screening is that positive predictive value is a function of prevalence, not merely of test accuracy. For a test with sensitivity Se, specificity Sp, and disease prevalence P:

$$
PPV = \frac{Se \times P}{Se \times P + (1 — Sp) \times (1 — P)}
$$

Consider a hypothetical assay with 50% sensitivity and 99.5% specificity:

At a prevalence of 0.1% — plausible for a single annual screening round in an asymptomatic population — more than nine in ten positive results would be false positives, even with a test exhibiting 99.5% specificity. This arithmetic explains why incremental improvements in specificity often matter more than equivalent improvements in sensitivity, and why published PPV values derived from enriched case-control cohorts cannot be transported to population screening.

6.4. Beyond Discrimination: The Metrics That Determine Utility

A clinically meaningful multi-omic platform must be evaluated on a multidimensional endpoint set:

• Sensitivity stratified by stage, particularly stage I-II.
• Specificity and false-positive rate in a prospectively enrolled asymptomatic cohort.
• Tissue-of-origin accuracy, which determines the efficiency of downstream workup.
• Lead time relative to clinical diagnosis.
• Overdiagnosis rate, i.e., detection of lesions that would never have caused symptoms.
• Diagnostic resolution burden, including number of imaging studies, invasive procedures, and time to resolution.
• Disease-specific and all-cause mortality reduction, the ultimate criterion.

7. Landmark Clinical Studies and the Current Evidence Base

7.1. CancerSEEK: Establishing the Multi-Analyte Principle

As discussed above, CancerSEEK demonstrated that combining mutant ctDNA with a protein panel yields substantially higher detection sensitivity than either modality alone. Its historical significance is conceptual. Its limitation is design: a retrospective case-control structure cannot estimate screening PPV or quantify overdiagnosis.

7.2. Targeted Methylation and the Galleri Assay

The most extensively validated multi-cancer early detection platform to date is the targeted methylation assay developed by GRAIL. In the published independent validation analysis, the assay achieved an overall sensitivity of approximately 51.5% at a specificity of 99.5%, with a marked stage dependence: sensitivity was approximately 16.8% for stage I disease and approximately 93.1% for stage IV. Cancer signal origin was predicted with approximately 88.9% accuracy.

These figures illustrate both the promise and the principal unresolved problem of the field. A test that detects the majority of stage III-IV cancers with very high specificity is analytically impressive. But a screening program whose purpose is to shift diagnosis toward curable stages must perform substantially better in stage I-II disease.

7.3. PATHFINDER: The Implementation Reality

The PATHFINDER study was a prospective implementation trial evaluating multi-cancer early detection testing in a community-based population of adults aged 50 years and older. It reported a positive predictive value of approximately 43.1% among participants with a cancer signal detected, together with substantial variation in the time and intensity of diagnostic resolution.

The study’s most important contribution was not a sensitivity estimate. It was the demonstration that a positive screening result generates a real clinical workflow: referral, imaging, endoscopy, biopsy, and, in a meaningful fraction of cases, no identifiable cancer. Any deployment model must budget for this cascade explicitly.

7.4. DELFI: Fragmentomics as a Complementary Signal

The DELFI framework demonstrated that genome-wide cfDNA fragmentation patterns alone can discriminate cancer patients from healthy individuals across several tumor types. Because fragmentation analysis can be derived from low-coverage sequencing or even from fragment-length distributions, it offers a potentially inexpensive orthogonal layer to combine with methylation and proteomic data.

7.5. UKCTOCS: A Cautionary Multimodal Precedent

The UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) is directly relevant to multi-omic thinking. It combined serial CA125 measurements with a risk-of-cancer algorithm — an early form of multivariate, longitudinally calibrated biomarker screening — in more than 200,000 postmenopausal women.

Despite improved detection characteristics relative to ultrasound-based screening, UKCTOCS did not demonstrate a statistically significant reduction in ovarian cancer mortality in its primary analysis. This outcome is a sobering reminder that improved stage shift and increased detection do not automatically translate into saved lives. The same standard must be applied to multi-omic blood tests.

8. Translational Barriers: Biology, Statistics, and Health Systems

8.1. The Shedding Problem

Not all tumors shed equably. Shedding depends on tumor volume, vascularization, anatomical location, necrotic fraction, and proximity to vascular or luminal compartments. Renal pelvis and bladder tumors may shed directly into urine; hepatocellular carcinomas shed abundantly into plasma; primary brain tumors are constrained by the blood-brain barrier and often release very little ctDNA.

A multi-omic assay cannot fully compensate for the absence of biological material. This is a fundamental physical constraint, not merely an engineering deficiency.

8.2. Overdiagnosis and the Lead-Time Illusion

Earlier detection can be harmful when it identifies indolent lesions that would never have become clinically apparent. Overdiagnosis subjects patients to anxiety, biopsy, surgery, and treatment-related morbidity without conferring survival benefit.

Two related biases must be controlled in trial design:

• Lead-time bias: survival appears longer simply because diagnosis occurred earlier, even if the time of death is unchanged.
• Length-time bias: screening preferentially detects slow-growing, less aggressive tumors, inflating apparent survival among screen-detected cases.

Only a randomized trial with a mortality endpoint can resolve these confounders.

8.3. Pre-Analytical Harmonization

Multi-omic assays are uniquely vulnerable to pre-analytical variation. Hemolysis contaminates cfDNA with germline leukocyte DNA; delayed centrifugation increases background cfDNA and degrades labile metabolites; repeated freeze-thaw cycles alter protein and metabolite profiles; collection tube chemistry influences both cellular lysis and metabolic stability.

A platform that performs exquisitely in a reference laboratory may fail in a distributed screening network unless pre-analytical conditions are standardized, audited, and enforced.

8.4. Equity, Cost, and Opportunity Cost

Multi-omic screening is expensive. Beyond the assay itself are the costs of diagnostic resolution, downstream surveillance, and treatment of screen-detected disease. Deployment must therefore be evaluated through formal health-economic modeling, and must explicitly address whether the technology narrows or widens existing disparities in cancer outcomes.

9. Clinical Positioning and the Regulatory Reality

A clear distinction must be maintained between three categories of liquid biopsy:

1. Therapy-selection companion diagnostics. Assays such as FoundationOne Liquid CDx and Guardant360 CDx are FDA-approved or cleared for identifying actionable alterations in patients with already-diagnosed advanced cancer. They are not screening tests.
2. Minimal residual disease and recurrence monitoring. Tumor-informed ctDNA assays detect molecular relapse after curative-intent treatment. This is a high-prevalence, high-pretest-probability setting in which performance differs fundamentally from screening.
3. Multi-cancer early detection in asymptomatic individuals. This is the most demanding setting and the one for which regulatory authorization remains unresolved.

As of the current regulatory field, no multi-cancer early detection blood test has been approved by the U.S. Food and Drug Administration for population screening. Professional guidance, including the interim clinical guidance issued by the American Cancer Society, recommends that MCED testing not replace established, evidence-based screening modalities and should be undertaken within informed, preferably research-oriented, frameworks.

The appropriate near-term clinical positioning is therefore enriched, high-risk populations: individuals with strong family history, known hereditary syndromes, chronic organ injury, or substantial tobacco exposure. Trials such as the NHS-Galleri study are designed to generate the population-level evidence that current case-control and implementation studies cannot provide.

Reporting standards must also be enforced. STARD 2015 governs diagnostic accuracy reporting; REMARK addresses tumor marker prognostic studies; TRIPOD governs multivariable prediction model development and validation. Adherence to these frameworks is a minimum requirement for credible translation.

10. Future Directions

Several convergent developments are likely to define the next generation of multi-omic detection.

Longitudinal personal baselines. Rather than comparing a patient to a population reference, a within-person trajectory — measured annually or biannually — may reveal subtle deviations that are invisible against interindividual biological variation. This is the principle that made serial CA125 algorithms more effective than single measurements in UKCTOCS, extended now to dozens of proteins and thousands of methylation loci.

Cross-layer foundation models. Large multimodal reference datasets may enable models that learn the joint manifold connecting genotype, chromatin state, protein expression, and metabolism. Such models could infer missing modalities, detect novel cancer types, and generate mechanistic hypotheses rather than purely statistical classifiers.

Single-cell and spatial contextualization. Multi-omic blood profiling will increasingly be interpreted alongside single-cell transcriptomic and spatial proteomic maps of the tumor microenvironment, allowing circulating signals to be traced back to their cellular sources.

Metabolic flux and isotope tracing. Moving beyond static metabolite concentrations toward measurements of pathway flux could substantially improve the specificity of the metabolic layer.

Closed-loop screening architectures. The most important systems-level innovation may not be a better assay but a better pathway: an integrated structure in which a positive blood test automatically triggers a standardized, risk-stratified, and time-limited diagnostic resolution protocol.

11. Conclusion

Multi-omic early cancer detection represents a genuine conceptual advance. Tumors disclose themselves through multiple, partially independent molecular channels, and the integration of circulating DNA, proteins, and metabolites can extract information that no single channel provides. The CancerSEEK and targeted methylation studies have established the analytical feasibility of this approach, while fragmentomics and large-scale plasma proteomics have expanded the available information space.

Nevertheless, the distance between analytical performance and clinical benefit remains substantial. Stage I sensitivity, prevalence-dependent predictive value, clonal hematopoiesis, overdiagnosis, pre-analytical variability, and diagnostic cascade burden are not peripheral technical issues; they are the central determinants of whether a screening program helps or harms.

The decisive evidence will therefore not come from a higher area under the curve in a retrospective cohort. It will come from prospectively enrolled, asymptomatic, externally validated populations followed long enough to measure stage shift, overdiagnosis, and mortality. Until such data exist, multi-omic liquid biopsy should be understood as a powerful and rapidly maturing research platform with defined adjunctive clinical roles — not as a validated replacement for the established architecture of cancer screening.

Sources and Key Research Centers

1. Johns Hopkins University, Sidney Kimmel Comprehensive Cancer Center — Victor Velculescu, Nickolas Papadopoulos, and colleagues. Development of CancerSEEK, ctDNA fragmentomics (DELFI), and tumor-informed minimal residual disease assays.
   • Cohen JD et al. «Detection and localization of surgically resectable cancers with a multi-analyte blood test.» Science, 2018; 359(6378): 926-930. DOI: 10.1126/science.aar3247
   • Cohen JD et al. «Combined circulating tumor DNA and protein biomarker-based liquid biopsy for the earlier detection of pancreatic cancers.» PNAS, 2017; 114(38): 10202-10207. DOI: 10.1073/pnas.1704961114
2. Stanford University, School of Medicine — Diehn / Alizadeh laboratories — Development of CAPP-Seq for ultrasensitive, personalized ctDNA detection.
   • Newman AM et al. «An ultrasensitive method for quantitating circulating tumor DNA with broad patient coverage.» Nature Biotechnology, 2016; 34(5): 547-555. DOI: 10.1038/nbt.3519
3. The Chinese University of Hong Kong — Dennis Lo laboratory — Foundational research on cell-free DNA biology, fragment length distributions, and non-invasive prenatal and cancer genomics.
   • Jiang P et al. «Lengthening and shortening of plasma DNA in hepatocellular carcinoma patients.» PNAS, 2015; 112(11): E1317-E1325. DOI: 10.1073/pnas.1500076112
4. Johns Hopkins University / Underhill, Levy and colleagues — DELFI fragmentomics — Genome-wide cfDNA fragmentation as a cancer detection modality.
   • Underhill HR et al. «Genome-wide cell-free DNA fragmentation in patients with cancer.» Nature, 2019; 570(7761): 385-389. DOI: 10.1038/s41586-019-1232-2
5. GRAIL, Inc. — Development and clinical validation of targeted methylation-based multi-cancer early detection.
   • Klein EA et al. «Clinical validation of a targeted methylation-based multi-cancer early detection test using an independent validation cohort.» Annals of Oncology, 2021; 32(9): 1167-1177. DOI: 10.1016/j.annonc.2021.05.806
6. Memorial Sloan Kettering Cancer Center — PATHFINDER prospective implementation study of multi-cancer early detection.
   • Schrag D et al. «Evaluating Multi-Cancer Early Detection Testing in a Community Setting: Primary Results From the PATHFINDER Study.» Journal of Clinical Oncology, 2023; 41(16): 3000-3009. DOI: 10.1200/JCO.22.02134
7. National Human Genome Research Institute / UK Biobank — Large-scale plasma proteomic phenotyping and genetic association studies.
   • Sun BB et al. «Plasma proteomic associations with genetics and health in the UK Biobank.» Nature, 2023; 622: 329-338. DOI: 10.1038/s41586-023-06592-7
8. Harvard Medical School / Broad Institute — Cantley and Thompson laboratories — Foundational mechanistic work on metabolic reprogramming in cancer.
   • Vander Heiden MG, Cantley LC, Thompson CB. «Understanding the Warburg effect: the metabolic requirements of cell proliferation.» Science, 2009; 324(5930): 1029-1033. DOI: 10.1126/science.1160809
   • DeBerardinis RJ et al. «The biology of cancer: metabolic reprogramming fuels cell growth and proliferation.» Cell Metabolism, 2008; 7(1): 11-20. DOI: 10.1016/j.cmet.2007.10.002
9. University of California, Los Angeles — Lusis laboratory — Conceptual framework for multi-omic integration in complex disease.
   • Hasin Y, Seldin M, Lusis A. «Multi-omics approaches to disease.» Genome Biology, 2017; 18: 83. DOI: 10.1186/s13059-017-1215-1
10. Queen Mary University of London / Barts Cancer Institute — UKCTOCS consortium — Randomized evaluation of multimodal ovarian cancer screening.
    • Jacobs IJ et al. «Ovarian cancer screening and mortality in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS): a randomised controlled trial.» The Lancet, 2016; 387(10022): 945-956. DOI: 10.1016/S0140-6736(15)01187-1
    • Menon U et al. «Risk algorithm using serial CA125 and HE4 concentrations to detect ovarian carcinoma in asymptomatic postmenopausal women.» Science Translational Medicine, 2009; 1(20): 20ra12. DOI: 10.1126/scitranslmed.3000510
11. National Cancer Institute — Early Detection Research Network (EDRN) — Federal program supporting biomarker discovery, assaying, and clinical validation for early cancer detection.
12. National Cancer Institute — The Cancer Genome Atlas (TCGA) — Foundational genomic and epigenomic characterization of human cancers underpinning target and biomarker selection.
13. American Cancer Society — Interim clinical guidance on multi-cancer early detection testing.
    • Wolf AMD et al. «Interim clinical guidance for screening with multicancer early detection tests.» CA: A Cancer Journal for Clinicians, 2022; 72(6): 521-532. DOI: 10.3322/caac.21744
14. U.S. Food and Drug Administration — Regulatory framework for liquid biopsy companion diagnostics and in vitro screening devices.
15. STARD / TRIPOD / REMARK reporting initiatives — Standards for diagnostic accuracy, prediction model, and biomarker prognostic study reporting.
    • Bossuyt PM et al. «STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies.» BMJ, 2015; 351: h5527. DOI: 10.1136/bmj.h5527
    • Collins GS et al. «Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement.» Annals of Internal Medicine, 2015; 162(1): 55-63. DOI: 10.7326/M14-0697
    • McShane LM et al. «REMARK: reporting recommendations for tumor marker prognostic studies.» Journal of the National Cancer Institute, 2005; 97(16): 1180-1184. DOI: 10.1093/jnci/dji237
16. ClinicalTrials.gov — Registry of ongoing multi-omic and multi-cancer early detection studies, including population-based MCED implementation trials. (Search terms: multi-cancer early detection, ctDNA screening, CancerSEEK, PATHFINDER, NHS-Galleri).