Functional Data Analysis in High-Dimensional Systems

Functional data analysis (FDA) extends classical statistics to the study of observations that vary over a continuum, such as curves, surfaces or images. In modern applications—from genomics and neuroimaging to finance and environmental modelling—data often reside in extremely high-dimensional spaces, challenging both computation and inference. FDA in high-dimensional systems seeks to represent complex functional observations in a lower-dimensional form while preserving salient temporal, spatial or spectral features. Core strategies involve the projection of infinite-dimensional objects onto finite bases, regularised estimation of covariance and regression operators, and manifold representations that respect underlying geometry. These methods enable the extraction of meaningful patterns, improve predictive accuracy and facilitate interpretation in settings where the number of functional measurements may far exceed the number of samples. As data acquisition technologies continue to advance, robust FDA frameworks for high-dimensional systems are becoming indispensable tools across scientific disciplines.

Research from Nature Portfolio

Recent work has advanced sparsity-driven modelling for longitudinal functional genomics by adapting the elastic net to time-course data. This approach jointly addresses high dimensionality and multicollinearity in gene expression profiles, selecting a sparse set of predictive trajectories across multiple time points. By combining penalised regression with time-series structure, the method identifies biomarkers whose temporal patterns predict treatment response with notably improved accuracy over conventional techniques. Computational innovations include efficient algorithms for tuning regularisation parameters and bootstrap frameworks for assessing selection stability. The resulting model provides a unified platform for prediction and biological interpretation in high-dimensional functional datasets.

Research from all publishers

A novel classification framework represents infinite-dimensional functional observations through a unified set of basis functions, transforming the original problem into a finite-dimensional parameter space. By optimising basis placements and employing B-spline constructions, the method achieves substantial gains in classification accuracy across diverse datasets. This unified representation simplifies computation while retaining essential functional structure, offering a versatile solution for non-linear classification tasks in infinite-dimensional spaces.

An interpretable discriminant analysis has been developed for functional data defined on random nonlinear domains, with applications in cortical surface geometry and thickness mapping. The approach recasts classification as a regularised multivariate functional linear model, estimating the most discriminant directions without pre-estimating the full covariance structure. Differential regularisation controls model complexity, yielding stable predictive features that align with known anatomical variations in neurodegenerative disease. This methodology enhances both interpretability and computational feasibility in high-dimensional manifold-valued functional data.

A manifold learning extension of stringing techniques maps high-dimensional observations onto functional representations by recovering an intrinsic ordering informed by data geometry. This ML-stringing approach replaces classical scaling with non-linear embeddings, producing smoother functional approximations that capture complex relationships among variables. Simulations and applications to gene expression arrays demonstrate that manifold-driven stringing can improve ordering quality and downstream analyses, opening new avenues for functional modelling of high-dimensional systems governed by underlying stochastic processes.

Functional Data Analysis in High-Dimensional Systems publication trend

The graph below shows the total number of articles in functional data analysis in high-dimensional systems across all publications each year (not limited to Nature Index journals).

Technical terms

Functional data: Observations viewed as functions over a continuum (for example, time, space or frequency).

High-dimensional systems: Settings in which the number of functional measurements or features greatly exceeds the number of samples.

Basis functions: A finite set of functions (such as splines or Fourier bases) used to represent infinite-dimensional objects in a tractable form.

Regularisation: The addition of penalty terms to estimation procedures to stabilise solutions in the presence of high dimensionality or multicollinearity.

Dimension reduction: Techniques for projecting high-dimensional functional observations onto a lower-dimensional subspace while preserving key variability.

Manifold learning: Unsupervised methods that uncover low-dimensional, non-linear structures embedded within high-dimensional data.

References

1. Classifying infinite-dimensional data with unified basis functions: An effective machine learning approach. Neurocomputing (2025).
2. Interpretable discriminant analysis for functional data supported on random nonlinear domains with an application to Alzheimer’s disease. Journal of the Royal Statistical Society Series B Statistical Methodology (2024).
3. Elastic net-based prediction of IFN-β treatment response of patients with multiple sclerosis using time series microarray gene expression profiles. Scientific Reports (2019).
4. Functional Modeling of High-Dimensional Data: A Manifold Learning Approach. Mathematics (2021).