Correction: Artificial intelligence in cancer epigenomics: a review on advances in pan-cancer detection and precision medicine

Abstract

Full text

Epigenetics & Chromatin (2025) 18:35

10.1186/s13072-025-00595-5

In Table 4 of this article, the title was inadvertently omitted.The correct title of Table 4 should be:

“Summary of DNA methylation-based MCED tests: key features and performance metrics from literature and industry sources”.

Table 4. Summary of DNA methylation-based MCED tests: key features and performance metrics from literature and industry sourcesCompany/ MethodTechnologyCancer types DetectedSensitivity/Specificity/TOO accuracySample TypeAI algorithmYear/CountryFeaturesRefs.IvyGeneWhole genome Methylation Analysis using PCR & NGSLiver, breast, colorectal, and lung cancers84%; 90%; N/APlasma (ctDNA)Utilizing AI, combined with cutting edge technology for analysing ctDNA methylation patterns.USA2018–2019.- Offers a non-invasive solution for early cancer detection.- Measures the status of the methylation levels of cfDNA.[124]cfMeDIP-SeqcfDNA Methylated DNA Immunoprecipitation SequencingPancreatic, colorectal, breast, lung, renal, bladder, AML, gliomaAUROC: 0.980 (AML), 0.918 (PDAC), 0.971 (LUC); High; N/APlasma (cfDNA)Incorporates ML (Random Forest) classifiers on cell-free DNA methylation.Canada; first major publication in 2019– 2020- Enriches for methylated DNA without bisulfite conversion.- Demonstrates high sensitivity to low fractions of cancer DNA.- Effective for subtype-specific and early-stage cancer detection.[125, 126]PanSEERctDNA-Methylation Biomarkers & PCR SequencingColorectal, esophageal, liver, lung, and stomach cancers88.2% (post-diagnosis);95%; N/APlasma (cfDNA)Utilizes a ML algorithm (logistic regression classifier) trained on methylation markers for early detection and TOO prediction.China/USA collaboration2020- Analyzes 477 cancer-specific DMRs with 10,613 CpGs,- Offers high sensitivity down to 0.01% cancer DNA.- Cost-effective with low cfDNA input.[127]GalleriTargeted DNA Methylation SequencingOver 50 cancer types51.5% overall; 16.8% (stage I), 40.4% (stage II), 77.0% (stage III), 90.1% (stage IV); 99.1%; 88.7%.Plasma (cfDNA)Advanced ML on targeted methylation data to detect and localize TOO of multiple cancers from cfDNA.USA; 2018.- High specificity for low tumor fraction.- Variable sensitivity by cancer type and stage.- Validated in CCGA and PATHFINDER trials.[128]CancerSEEK/Exact Science (USA)Mutation and Protein Biomarker Analysis; Liquid Biopsy using cfDNA and ML for cancer predictionAbout 8 cancer types(ovary, liver, stomach)62% (1005 as number of patient with cancer); 99.1%; 63%.Plasma (ctDNA)ML classifier combining ctDNA mutations and selected protein biomarkers.USA;2019–2020.- Integrates multi-omics.- Detects cancers without standard screening.[123, 129]Adela Bio/AdelaCell-free methylated DNA immunoprecipitation-sequencing.About 7 cancer types70%–75%; 99%; 91%.Plasma (cfDNA)ML application analyzing methylome signalsUSA; 2021- Avoids bisulfite conversion;- Effective for early-stage and subtype-specific detection;- Utilizes ML algorithms to identify key methylated regions in cfDNA.[130]TR(ACE) /Biological dynamicsAlternating current electrokinetics (ACE) platform to purify extracellular vesicles from plasma; ML algorithm.About 7 cancer typesSensitivity of 71.2% (95% CI: 63.2–78.1); 99.5% ; 43.8% - 95.5%Circulating extracellular vehicles (EVs)Uses ML classifier on extracellular vesicles (EV)–associated biomarkers.USA;2020–2021- Utilizes ACE platform to isolate circulating EVs from plasma and is used for multi-marker analysis.- Efficient analysis of a large number of samples.[131]CancerdetectorcfDNA bisulfite sequencing, probabilistic model.Study report on liver cancer but claims to detect all types of cancers.94.8%; 100%; N/A.Plasma (cfDNA)NAUSA;2018- Developed CancerDetector which focuses on joint methylation states improves sensitivity for detecting abnormal cfDNAs.- Achieves high accuracy and its prediction is consistent with clinical information.[132]EpiPanGI DxBisulfite sequencing method and Machine learningGI cancers (CRC, pancreatic, stomach)AUC: 0.88; 96%; 0.85–0.95.Plasma (cfDNA)Methylation-based biomarker approach with machine learning for GI cancer detection.U.S., Germany, Japan, South Africa, and Spain; 2020–2021.- This test identified the three distinct DMR panels that are Cancer-Specific Biomarker Panels.- Pan-GI Cancer Panel and multi-cancer TOO prediction panel.[133]Burning Rock DxTargeted methylation sequencing assay combined with machine learning.6 Cancers (liver, colon/rectum, esophagus, pancreas, lung and ovary)80.6%; 98.3%; 81.0%.Plasma (cfDNA)- ML approaches (SVM) for ctDNA methylation or targeted gene panel detection.- Multi-class logistic regression was used to predict tissue origin.China; 2020- Optimized for low-depth sequencing; validated in THUNDER-II trial.- Shows high specificity and accurate TOO prediction.[134]GENECASTTargeted methylation sequencing14 cancer types72.86% ; 96.67% (AUC = 0.86); N/A.Plasma (cfDNA)N/AChina;2019–2021- The model developed was based on 37 MCB. Biomarker methylation differences were computed using a HM-score.[135]Guardant RevealMethylation panel (500 CpGs) + fragmentomics + ML13 cancers76.4%/42%; 97.9%; 82%.ctDNAStatistical and ML–based algorithm for ctDNA signals (both genomic, epigenomic and fragmentomics).Usa; 2021- Combines methylation and fragment size analysis.- FDA-approved for colorectal cancer recurrence.[136]DELFI (Delfi DX)Genome-wide cfDNA fragmentation + ML7 cancers (lung, breast, liver)73% (Stage I–II); 98%; 85%.Plasma (cfDNA)ML on cfDNA fragmentomics (fragmentation profiles).USA; 2019- Low-cost WGS approach.- Detects fragmentation patterns linked to chromatin instability.[137]OncoSeekProtein biomarkers (CA-125, CEA) + AI9 cancers51.7%; 95%; 66.8%.Plasma (serum)Uses AI for calculating the POC indexCalifornia, USA;- Low-cost protein-based test.- Integrates clinical metadata for risk stratification.[138]SPOT-MAS (Gene Solutions)Targeted methylation (14 genes) + ML5 cancers (breast, liver, CRC, lung)78%; 99.8%; 84%ctDNAIntegrates ML analysis on ctDNA mutation and methylation signals.Vietnam; 2021–2022- Validated in 10,000+ Vietnamese patients.- Optimized for low-resource settings.[139, 140]SeekInCare (SeekIn)Methylation + CNVs + ML20 cancers (NSCLC, CRC, liver)65.5%; 97%; 93%Plasma (cfDNA)Deals with Multi-omic (ctDNA + protein) data; gradient-boosting machine ML algorithms for early detection and surveillance.China; 2020–2021.- Resource-optimized cancer screening for large populations.[141]FreenomeMethylation + fragmentomics + proteomics + gradient boosting8 cancers (CRC, lung, breast)79.2%; 92% in CRC; N/A.Plasma (cfDNA)ML enabled multi-omicsUSA; 2019- Multi-omics approach;- This is under clinical evaluation in PROSPECT study (NCT05581476).[142]*a. ACE: alternating current electrokinetics, b. EV: Extracellular vesicles, c. MCB: methylation-correlated blocks, d. HM score: Hypermethylation score; e. WGS: Whole genome sequencing. f. SVM: Support Vector Machine., g. POC: probability of cancer; h. ML: Machine learning, i. AI: Artificial Intelligence.; f. N/A- Not available

The original article has been corrected.