{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T19:00:03Z","timestamp":1785438003905,"version":"3.56.0"},"posted":{"date-parts":[[2026,6,18]]},"group-title":"Genetic and Genomic Medicine","reference-count":25,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T00:00:00Z","timestamp":1781740800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"funder":[{"award":["P30 CA023108"],"award-info":[{"award-number":["P30 CA023108"]}],"id":[{"id":"https:\/\/ror.org\/001d55x84","id-type":"ROR","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,6,18]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>DNA methylation-based classification offers a rapid diagnostic complement to conventional molecular workflows in acute leukemia. Existing classifiers are trained on array-derived reference cohorts whose construction favors specimens with adequate tumor content, leaving clinically relevant low-purity specimens underrepresented and classifier robustness in this regime uncharacterized. On held-out low-purity specimens, existing classifiers were concordant with expert pathology in only 7 of 10 (MARLIN) and 5 of 10 (ALMA) cases, motivating a classifier built to maintain accuracy at low tumor purity. We developed MOSAIC (Methylation-Oriented Site Analysis and Information Classifier), a neural network classifier built to maintain accuracy across the full range of tumor purities encountered in clinical practice. MOSAIC is a neural network trained on publicly available array-based methylation data augmented with native methylation calls from Oxford Nanopore sequencing. MOSAIC was evaluated on low-purity specimens held out entirely from training. On these held-out low-blast leukemia specimens, all below 25% blasts and including a case at 1.4%, MOSAIC was concordant with expert pathology in every case, recovering the correct subtype where diluted disease signal would otherwise be mistaken for normal or unrelated tissue. Gradient-based saliency analysis showed that the network relies on a partially distinct set of discriminative CpG probes when classifying low-blast specimens. MOSAIC demonstrates that augmenting training with clinically representative clinical specimens yields methylation-based leukemia classification that maintains effectiveness under the variable tumor purity of real clinical cohorts.<\/jats:p>","DOI":"10.64898\/2026.06.16.26355747","type":"posted-content","created":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T23:20:13Z","timestamp":1781824813000},"source":"Crossref","is-referenced-by-count":0,"title":["MOSAIC: Methylation-Oriented Site Analysis and Information Classifier for Robust Epigenomic Classification of Acute Leukemia in Clinical Cohorts with Variable Tumor Purity"],"prefix":"10.64898","author":[{"given":"Atharva Jaydeep","family":"Shah","sequence":"first","affiliation":[{"name":"Indiana University, Bloomington, IN, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Donald","family":"Green","sequence":"additional","affiliation":[{"name":"Dartmouth-Hitchcock Medical Center, Lebanon, NH, 03766, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lauren","family":"Wainman","sequence":"additional","affiliation":[{"name":"Dartmouth-Hitchcock Medical Center, Lebanon, NH, 03766, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jeremiah","family":"Karrs","sequence":"additional","affiliation":[{"name":"Dartmouth-Hitchcock Medical Center, Lebanon, NH, 03766, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Parth","family":"Shah","sequence":"additional","affiliation":[{"name":"Dartmouth-Hitchcock Medical Center, Lebanon, NH, 03766, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026062119300436000_2026.06.16.26355747v1.1","doi-asserted-by":"publisher","DOI":"10.1182\/blood2022016867"},{"key":"2026062119300436000_2026.06.16.26355747v1.2","doi-asserted-by":"publisher","DOI":"10.1038\/s41375-022-01613-1"},{"key":"2026062119300436000_2026.06.16.26355747v1.3","doi-asserted-by":"publisher","DOI":"10.1182\/BLOOD.2022015850"},{"key":"2026062119300436000_2026.06.16.26355747v1.4","doi-asserted-by":"crossref","first-page":"e70083","DOI":"10.1002\/hem3.70083","article-title":"Diagnosis and treatment of AML in the context of WHO and ICC 2022 classifications: Divergent nomenclature converges on common therapies","volume":"9","year":"2025","journal-title":"HemaSphere"},{"key":"2026062119300436000_2026.06.16.26355747v1.5","doi-asserted-by":"publisher","DOI":"10.1182\/blood.2022015853"},{"key":"2026062119300436000_2026.06.16.26355747v1.6","doi-asserted-by":"publisher","DOI":"10.1038\/nature26000"},{"key":"2026062119300436000_2026.06.16.26355747v1.7","doi-asserted-by":"publisher","DOI":"10.1093\/neuonc\/noab106"},{"key":"2026062119300436000_2026.06.16.26355747v1.8","doi-asserted-by":"publisher","DOI":"10.1007\/s00401-017-1743-5"},{"key":"2026062119300436000_2026.06.16.26355747v1.9","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06615-2"},{"key":"2026062119300436000_2026.06.16.26355747v1.10","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-025-02321-z"},{"key":"2026062119300436000_2026.06.16.26355747v1.11","doi-asserted-by":"publisher","DOI":"10.21203\/rs.3.rs-5450972\/v1"},{"key":"2026062119300436000_2026.06.16.26355747v1.12","doi-asserted-by":"publisher","DOI":"10.1182\/bloodadvances.2024015275"},{"key":"2026062119300436000_2026.06.16.26355747v1.13","doi-asserted-by":"crossref","first-page":"e0265557","DOI":"10.1371\/journal.pone.0265557","article-title":"Tumor purity adjusted beta values improve biological interpretability of high-dimensional DNA methylation data","volume":"17","year":"2022","journal-title":"PLOS ONE"},{"key":"2026062119300436000_2026.06.16.26355747v1.14","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btx303"},{"key":"2026062119300436000_2026.06.16.26355747v1.15","doi-asserted-by":"publisher","DOI":"10.1186\/s12859-019-3014-z"},{"key":"2026062119300436000_2026.06.16.26355747v1.16","doi-asserted-by":"crossref","first-page":"80","DOI":"10.1038\/s41375-021-01316-z","article-title":"Investigation of measurable residual disease in acute myeloid leukemia by DNA methylation patterns","volume":"36","year":"2022","journal-title":"Leukemia"},{"key":"2026062119300436000_2026.06.16.26355747v1.17","doi-asserted-by":"crossref","unstructured":"Quinonero-Candela, J. , Sugiyama, M. , Schwaighofer, A. & Lawrence, N. D. Dataset Shift in Machine Learning (MIT Press, 2008).","DOI":"10.7551\/mitpress\/9780262170055.001.0001"},{"key":"2026062119300436000_2026.06.16.26355747v1.18","doi-asserted-by":"publisher","DOI":"10.1056\/NEJMc2104626"},{"key":"2026062119300436000_2026.06.16.26355747v1.19","first-page":"124003","article-title":"Deep double descent: where bigger models and more data can hurt","year":"2021","journal-title":"J. Stat. Mech"},{"key":"2026062119300436000_2026.06.16.26355747v1.20","first-page":"3912","article-title":"Investigating the impact of model width and density on generalization in presence of label noise","volume":"244","year":"2024","journal-title":"Proc. 40th Conf. on Uncertainty in Artificial Intelligence (UAI), PMLR"},{"key":"2026062119300436000_2026.06.16.26355747v1.21","unstructured":"Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization (2015). URL https:\/\/arxiv.org\/abs\/1412.6980."},{"key":"2026062119300436000_2026.06.16.26355747v1.22","article-title":"Deep inside convolutional networks: visualising image classification models and saliency maps","year":"2014","journal-title":"Proc. 2nd International Conference on Learning Representations (ICLR), Workshop Track"},{"key":"2026062119300436000_2026.06.16.26355747v1.23","doi-asserted-by":"crossref","unstructured":"Subasri, V. , Krishnan, A. , Dhalla, A. et al. Diagnosing and remediating harmful data shifts for the responsible deployment of clinical AI models. medRxiv (2023).","DOI":"10.1101\/2023.03.26.23286718"},{"key":"2026062119300436000_2026.06.16.26355747v1.24","doi-asserted-by":"crossref","first-page":"1249258","DOI":"10.3389\/fdgth.2023.1249258","article-title":"Assessing optimal methods for transferring machine learning models to low-volume and imbalanced clinical datasets: experiences from predicting outcomes of Danish trauma patients","volume":"5","year":"2023","journal-title":"Front. Digit. Health"},{"key":"2026062119300436000_2026.06.16.26355747v1.25","doi-asserted-by":"crossref","first-page":"8380","DOI":"10.1038\/s41598-022-12497-7","article-title":"Leveraging clinical data across healthcare institutions for continual learning of predictive risk models","volume":"12","year":"2022","journal-title":"Sci. Rep"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.06.16.26355747","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,22]],"date-time":"2026-06-22T02:30:23Z","timestamp":1782095423000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.64898\/2026.06.16.26355747"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6,18]]},"references-count":25,"URL":"https:\/\/doi.org\/10.64898\/2026.06.16.26355747","relation":{},"subject":[],"published":{"date-parts":[[2026,6,18]]},"subtype":"preprint"}}