Detecting Batch Heterogeneity via Likelihood Clustering

TL;DR

Batch effects in CNV detection from NGS confound references, risking false CNV calls or missed pathogenic variants. The paper proposes clustering samples by Bayesian model evidence $\ell_i$ derived from two CNV generative models and uses a parametric bootstrap–based likelihood-ratio test to detect mixture structure in evidence space. It validates the approach with synthetic data and three clinical panels plus mouse electrophysiology, showing improved clustering accuracy over conventional correlation-based and dimensionality-reduction baselines while maintaining conservative false-positive control. This framework enables label-free, robust detection of batch heterogeneity, guiding process-matched sample selection and enhancing clinical reliability in CNV analysis.

Abstract

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.

Figures (4)

• Figure 1: Overview of evidence computation. Left: a sequence of power posteriors indexed by inverse temperature $\tau \in [0,1]$ used for thermodynamic integration. Second from left: thermodynamic integration estimates $\log p(y)$ by numerically integrating $\mathbb{E}_{p_{\tau}(\theta\mid y)}[\log p(y\mid \theta)]$ over $\tau$ (Eq. \ref{['eq:ti_property']}). Middle right: particle-filter weights at a representative amplicon position; highly peaked weight distributions indicate that few particles explain the observation well. Right: particle-filter evidence accumulates across amplicons via the product form in Eq. \ref{['eq:pf_evidence']}.
• Figure 2: Overview of the proposed heterogeneity detection workflow. On the top left we show two sample LCNrs after preprocessing according to \ref{['eq:centering']}. The top right shows an example computation of the evidence, matching the right of Figure \ref{['fig:evidence']}. Bottom left shows the distribution of the evidences of the samples in the entire batch with a clear bimodal structure and color corresponding to group label. Finally, bottom right visualizes the significance test defined by Algorithm \ref{['alg:bootstrap_lrt']}.
• Figure 3: The left figure shows statistical power as a function of sample size in each of the three CNV assays. The right figure shows the distributions of p-values at given sample sizes for the thalassemia and LBX assays.
• Figure 4: (Left) A histogram of observed p-values (blue) with the theoretical uniform density (dashed gray line) and the fitted $\mathrm{Beta}(\hat{a}, 1)$ distribution overlaid (red) using the gennorm kernel. The shaded region represents the 95% credible interval derived from the posterior uncertainty in $\hat{a}$. (Right) A Q--Q plot comparing observed p-values to expected quantiles from a uniform distribution (dashed gray line) and from the fitted $\mathrm{Beta}(\hat{a}, 1)$ model (red). The gray shaded region denotes the 95% confidence band for order statistics under the null. Systematic deviations from the uniform reference line indicate a departure from the null distribution, consistent with the fitted model.