Using flexible noise models to avoid noise model misspecification in inference of differential equation time series models

When modelling time series, it is common to decompose observed variation into a "signal" process, the process of interest, and "noise", representing nuisance factors that obfuscate the signal. To separate signal from noise, assumptions must be made about both parts of the system. If the signal process is incorrectly specified, our predictions using this model may generalise poorly; similarly, if the noise process is incorrectly specified, we can attribute too much or too little observed variation to the signal. With little justification, independent Gaussian noise is typically chosen, which defines a statistical model that is simple to implement but often misstates system uncertainty and may underestimate error autocorrelation. There are a range of alternative noise processes available but, in practice, none of these may be entirely appropriate, as actual noise may be better characterised as a time-varying mixture of these various types. Here, we consider systems where the signal is modelled with ordinary differential equations and present classes of flexible noise processes that adapt to a system's characteristics. Our noise models include a multivariate normal kernel where Gaussian processes allow for non-stationary persistence and variance, and nonparametric Bayesian models that partition time series into distinct blocks of separate noise structures. Across the scenarios we consider, these noise processes faithfully reproduce true system uncertainty: that is, parameter estimate uncertainty when doing inference using the correct noise model. The models themselves and the methods for fitting them are scalable to large datasets and could help to ensure more appropriate quantification of uncertainty in a host of time series models.

[1]  David B. Dahl,et al.  Sequentially-Allocated Merge-Split Sampler for Conjugate and Nonconjugate Dirichlet Process Mixture Models , 2005 .

[2]  Ajay Jasra,et al.  On population-based simulation for static inference , 2007, Stat. Comput..

[3]  Jianqing Fan,et al.  Sparsistency and Rates of Convergence in Large Covariance Matrix Estimation. , 2007, Annals of statistics.

[4]  M. Wand,et al.  Simple Marginally Noninformative Prior Distributions for Covariance Matrices , 2013 .

[5]  P. Bickel,et al.  Regularized estimation of large covariance matrices , 2008, 0803.1909.

[6]  David A. Landgrebe,et al.  Covariance Matrix Estimation and Classification With Limited Training Data , 1996, IEEE Trans. Pattern Anal. Mach. Intell..

[7]  Mark J. Schervish,et al.  Nonstationary Covariance Functions for Gaussian Process Regression , 2003, NIPS.

[8]  Søren Hauberg,et al.  Geodesic exponential kernels: When curvature and linearity conflict , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[9]  G. Fasshauer Positive definite kernels: past, present and future , 2011 .

[10]  Dorota Kurowicka,et al.  Generating random correlation matrices based on vines and extended onion method , 2009, J. Multivar. Anal..

[11]  P J Diggle,et al.  Nonparametric estimation of covariance structure in longitudinal data. , 1998, Biometrics.

[12]  Juho Rousu,et al.  Non-Stationary Gaussian Process Regression with Hamiltonian Monte Carlo , 2015, AISTATS.

[13]  Jorge Nocedal,et al.  Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization , 1997, TOMS.

[14]  Pras Pathmanathan,et al.  Uncertainty and variability in models of the cardiac action potential: Can we build trustworthy models? , 2016, Journal of molecular and cellular cardiology.

[15]  Diane J. Cook,et al.  A survey of methods for time series change point detection , 2017, Knowledge and Information Systems.

[16]  Michael Clerx,et al.  Accounting for variability in ion current recordings using a mathematical model of artefacts in voltage-clamp experiments , 2019, bioRxiv.

[17]  H. Haario,et al.  An adaptive Metropolis algorithm , 2001 .

[18]  J. Milstein,et al.  The Inverse Problem: Estimation of Kinetic Parameters , 1981 .

[19]  Christopher J. Paciorek,et al.  Nonstationary Gaussian Processes for Regression and Spatial Modelling , 2003 .

[20]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.

[21]  Tina Toni,et al.  Designing attractive models via automated identification of chaotic and oscillatory dynamical regimes , 2011, Nature communications.

[22]  E FasshauerG Positive definite kernels: past, present and future , 2011 .

[23]  Chon Lok Lei,et al.  Rapid Characterization of hERG Channel Kinetics I: Using an Automated High-Throughput System , 2019, Biophysical journal.

[24]  Gary R. Mirams,et al.  Rapid Characterization of hERG Channel Kinetics II: Temperature Dependence , 2019, bioRxiv.

[25]  John K Kruschke,et al.  Bayesian data analysis. , 2010, Wiley interdisciplinary reviews. Cognitive science.

[26]  Ramsés H. Mena,et al.  On a Nonparametric Change Point Detection Model in Markovian Regimes , 2014 .

[27]  Weidong Liu,et al.  Adaptive Thresholding for Sparse Covariance Matrix Estimation , 2011, 1102.2237.

[28]  K. Strimmer,et al.  Statistical Applications in Genetics and Molecular Biology A Shrinkage Approach to Large-Scale Covariance Matrix Estimation and Implications for Functional Genomics , 2011 .

[29]  A. Hodgkin,et al.  A quantitative description of membrane current and its application to conduction and excitation in nerve , 1952, The Journal of physiology.

[30]  Norbert Wiener,et al.  Extrapolation, Interpolation, and Smoothing of Stationary Time Series, with Engineering Applications , 1949 .

[31]  Radford M. Neal MCMC Using Hamiltonian Dynamics , 2011, 1206.1901.

[32]  Andrew Gelman,et al.  Handbook of Markov Chain Monte Carlo , 2011 .

[33]  David Higdon,et al.  Non-Stationary Spatial Modeling , 2022, 2212.08043.

[34]  Carmen G. Moles,et al.  Parameter estimation in biochemical pathways: a comparison of global optimization methods. , 2003, Genome research.