A nonparametric calibration of a pseudo-LRT for human circadian gene expression detection

Y. Larriba, I. Fernández, J. S. Abston

We propose a framework for circadian rhythmicity detection in transcriptomic data with irregular circadian-time sampling and low signal-to-noise ratio based on the single-component Frequency Modulated Möbius model (FMM₁). The model provides a flexible representation of oscillatory signals on the circular domain with physiologically interpretable parameters. Rhythmicity is defined through the wave-sharpness parameter ω, testing a flat signal (H₀: ω = 0) against a rhythmic alternative (H₁: ω > 0). The null model involves both a boundary constraint and loss of identifiability, resulting in a nonstandard distribution for the test statistic. We address this using a weighted pseudo-LRT, calibrated via wild bootstrap and sequential Monte Carlo. Estimation is stabilized through inverse-density weighting and curvature regularization. Simulations show good type-I error control and improved power for non-sinusoidal signals. Application to GTEx data reveals novel circadian genes in human tissues.

Keywords: FMM model, Circular data, Nonparametric inference, Bootstrap, Likelihood ratio test, Circadian rhythms

GT Estadística no Paramétrica III: Inferencia no paramétrica para datos circulares
September 5, 2026  10:00 AM
Aula 29

A. Fernández de Marcos, E. García Portugués

A. Meilán Vila, M. Francisco Fernández

D. Bolón Rodríguez, D. Paindaveine