Transforming blood-derived episignatures into cell-type-agnostic classifiers: A shortcut to prenatal episignatures.

DNA methylation episignatures converted into cell-type-agnostic classifiers by machine learning, in order to overcome the tissue specificity of blood-derived signatures.

DNA methylation episignatures are valuable biomarkers for assessing variant pathogenicity in neurodevelopmental disorders, but whole blood-derived signatures are tissue- and cell-type specific, limiting their use in prenatal diagnostics. The authors ran a proof-of-concept study on trisomy 21, generating a blood-derived episignature from 266 samples and then training machine-learning models on 850 publicly available trisomy 21 and control samples across six pre- and postnatal tissues. Models trained on postnatal blood-derived signatures, as well as on other tissues, accurately predicted trisomy 21 status across all tested tissues, yielding disease-specific yet cell-type-agnostic patterns. Combining well-characterised postnatal episignatures with a limited set of prenatal samples was enough to generate a signature that correctly classified prenatal samples.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Tissue dependency is the main obstacle to prenatal use of episignatures, and showing it can be circumvented without assembling large prenatal cohorts is a meaningful methodological result. The caveat concerns the chosen model: trisomy 21 is a global dosage signal, far stronger than the episignature of an isolated missense variant in a chromatin disorder — transferability to the cases that actually need it remains to be shown. Worth following as a framework, not yet as a prenatal test.

Analysis by Dr Thibaut Benquey