Genetic diagnostics is like comparing two photos taken with different cameras. If one camera slightly shifts the colors, the image will show "defects" that don't exist in reality. The same happens in DNA analysis: a patient's sample is compared to a reference, and if the reference was processed with reagents from a different batch or in a different lab, the comparison yields false signals — either extra mutations or missed real ones.
The authors propose catching such glitches by the "level of trust" between the data and the model. When the technique works as intended, the data fit the predictions well. When a reagent batch is spoiled or the instrument is miscalibrated, the data stop converging with the model — trust drops. The method looks for groups of samples with abnormally low trust and separates them from healthy ones. It's like finding photos with unnatural colors in an album: they stand out against the rest.
Testing on synthetic data showed that the method doesn't raise false alarms more often than expected. On real clinical panels (liquid biopsy, BRCA, and thalassemia) it found glitches caused by various reasons, from reagent changes to processing in different locations. And on mouse brain electrical activity recordings, the method worked just as well, even though the task is from a completely different field. This means the approach is universal: it catches any technical interference that spoils comparison with a reference.
Such a filter isn't just for show. In the clinic, a false mutation can send a person into unnecessary treatment, while a missed real one can leave them without help. The method allows sorting out spoiled tests before they reach the doctor's hands, thereby maintaining trust in genetic diagnostics.
as written by the authors arXiv:2601.09758
Batch effects represent a major confounder in genomic diagnostics. In copy number variant (CNV) detection from NGS, many algorithms compare read depth between test samples and a reference sample, assuming they are process-matched. When this assumption is violated, with causes ranging from reagent lot changes to multi-site processing, the reference becomes inappropriate, introducing false CNV calls or masking true pathogenic variants. Detecting such heterogeneity before downstream analysis is critical for reliable clinical interpretation. Existing batch effect detection methods either cluster samples based on raw features, risking conflation of biological signal with technical variation, or require known batch labels that are frequently unavailable. We introduce a method that addresses both limitations by clustering samples according to their Bayesian model evidence. The central insight is that evidence quantifies compatibility between data and model assumptions, technical artifacts violate assumptions and reduce evidence, whereas biological variation, including CNV status, is anticipated by the model and yields high evidence. This asymmetry provides a discriminative signal that separates batch effects from biology. We formalize heterogeneity detection as a likelihood ratio test for mixture structure in evidence space, using parametric bootstrap calibration to ensure conservative false positive rates. We validate our approach on synthetic data demonstrating proper Type I error control, three clinical targeted sequencing panels (liquid biopsy, BRCA, and thalassemia) exhibiting distinct batch effect mechanisms, and mouse electrophysiology recordings demonstrating cross-modality generalization. Our method achieves superior clustering accuracy compared to standard correlation-based and dimensionality-reduction approaches while maintaining the conservativeness required for clinical usage.
The method was tested on mouse brain activity recordings, and it found technical glitches in the data, even though it was created for DNA analysis. It's like a counterfeit bill detector suddenly recognizing a forgery in a painting.