Abstract
<title>Abstract</title> <p>Background. Detecting hematologic malignancy before clinical presentation would convert a fixed-outcome diagnosis into a modifiable risk state. DNA methylation is the earliest measurable record of hematopoietic perturbation, and pre-diagnostic cohorts show genome-wide deviation from healthy baselines months to years before diagnosis. Two distinct questions follow: whether an individual's current methylation state is anomalous (a classification question), and whether that state is moving toward disease with a measurable velocity and lead time (a trajectory question). The clinical value lies in the second; the scientific hazard is that trajectory language is easy to assert and hard to earn. Methods. Using only publicly deposited Illumina HumanMethylation450/EPIC cohorts, we quantified static methylation burden (mean absolute β-deviation from a healthy whole-blood reference) and regulatory entropy (Shannon entropy of the deviation distribution), and evaluated a locked directional 48-CpG burden screen (qMethyl-48; locus composition withheld pending intellectual-property prosecution) against public healthy and disease cohorts. No proprietary, electronic-health-record, or individual patient-record data were used in this manuscript. Results. Static burden and entropy separate WHO-2022 hematologic-malignancy subtypes from healthy donors and rank subtypes coherently. Against 656 healthy whole-blood donors, the qMethyl-48 screen attained 96.6% specificity at its locked threshold, with 94.9–100% sensitivity across chronic myeloid leukemia, chronic myelomonocytic leukemia, and myeloproliferative neoplasms. All results are single-timepoint classification measures. Interpretation. These static signals do not license per-patient trajectory inference. Velocity, arrival, and conservation are properties of a dynamical system observed over time; cross-sectional data contains no time axis and two-point data the degenerate minimum, so a rate computed as a static deviation divided by an assumed lead time encodes no measured derivative and is circular by construction. We frame trajectory recovery as a state-space-reconstruction problem, state the identifiability question precisely, and specify a prospective serial-sampling program with pre-specified go/no-go gates capable of substantiating or refuting trajectory-based lead-time estimation. The work separates what is usable now, static risk stratification, deployable as a laboratory-developed test from what must be prospectively earned.</p>