Abstract:BACKGROUND:Quantifying associations in neuroscience (and many other scientific disciplines) is often challenged by high-dimensionality, nonlinearity and noisy observations. Many classic methods have either poor power or poor scalability on data sets of the same or different scales such as genetical, physiological and image data. NEW METHOD:Based on the framework of reproducing kernel Hilbert spaces we proposed a new nonlinear association criteria (NAC) with an efficient numerical algorithm and p-value approximation scheme. We also presented mathematical justification that links the proposed method to related methods such as kernel generalized variance, kernel canonical correlation analysis and Hilbert-Schmidt independence criteria. NAC allows the detection of association between arbitrary input domain as long as a characteristic kernel is defined. A MATLAB package was provided to facilitate applications. RESULTS:Extensive simulation examples and four real world neuroscience examples including functional MRI causality, Calcium imaging and imaging genetic studies on autism [Brain, 138(5):13821393 (2015)] and alcohol addiction [PNAS, 112(30):E4085-E4093 (2015)] are used to benchmark NAC. It demonstrates the superior performance over the existing procedures we tested and also yields biologically significant results for the real world examples. COMPARISON WITH EXISTING METHOD(S):NAC beats its linear counterparts when nonlinearity is presented in the data. It also shows more robustness against different experimental setups compared with its nonlinear counterparts. CONCLUSIONS:In this work we presented a new and robust statistical approach NAC for measuring associations. It could serve as an interesting alternative to the existing methods for datasets where nonlinearity and other confounding factors are present.

Effects of statistical distribution on nonlinear correlation coefficient

A new non-liner correlation measure

Correlation Functions of an Autonomous Stochastic System with Nonlinear Time Delays

Study on Linear Correlation Coe Cient and Nonlinear Correlation Coe Cient in Mathematical Statistics

A Nonlinear Correlation Measure for Multivariable Data Set

Reconstruction of Noise-Driven Nonlinear Networks from Node Outputs by Using High-Order Correlations

ASYMPTOTIC DISTRIBUTIONS OF HIGH-DIMENSIONAL DISTANCE CORRELATION INFERENCE

An improved nonlinear correlation method for feature selection of complex data

Study on the performance of the nonlinear joint transform correlator

Nonlinear association criterion, nonlinear Granger causality and related issues with applications to neuroimage studies.

Systematically Exploring Associations among Multivariate Data

A note on measuring the non-linear dependence via sub-copula based regression

Local Influence In Nonlinear Reproductive Dispersion Models

A Novel Kernel Correlation Coefficient with Robustness Against Nonlinear Attenuation and Impulsive Noise

Enlarged Scaling Ranges and Smooth Boundary for Detection of Chaos

Adjust Pearson's r to Measure Nonlinear Monotone Dependence

Nonlinear statistical coupling

Effects of Nonlinear Time-Delay on A Stochastic Asymmetric System

Network Maximal Correlation

With random regressors, least squares inference is robust to correlated errors with unknown correlation structure

The effects of degree correlations on network topologies and robustness