Analysis of brain connectivity in autism spectrum disorder using large-scale non-linear Granger causality (lsNGC)

We introduce large-scale non-linear Granger causality (lsNGC) as a novel method for analyzing directed connectivity in high-dimensional neural systems. The lsNGC algorithm combines invertible dimension reduction with non-linear time-series prediction, addressing limitations of existing approaches in connectivity analysis of complex systems. This method is multivariate, capable of identifying the influence of each time-series on any other in the presence of all other time-series of the underlying dynamic system. We apply lsNGC to resting-state fMRI data from the Autism Brain Imaging Data Exchange II (ABIDE II) dataset, specifically using data from the Olin Neuropsychiatry Research Center. Our method demonstrates superior performance in classifying individuals with autism spectrum disorder (ASD) versus healthy controls (HLT). Quantitative evaluation shows that lsNGC achieves a higher AUC of 0.935 ± 0.076 compared to competing methods: Pearson correlation (0.890 ± 0.100, −4.81%) and antisymmetric correlation (0.882 ± 0.102, −5.67%). Statistical analysis using Mann-Whitney U tests confirms significant improvements of lsNGC over all other competting methods, namely, Pearson correlation, local models, and antisymmetric correlation (p ⪅ 0.003). These results indicate lsNGC’s potential for enhancing our understanding of brain connectivity patterns in ASD and other neurological conditions. By extracting sparse connectivity matrices, lsNGC may be particularly useful for network analysis in complex systems and could be applicable to clinical fMRI analysis in future research, such as targeting disease-related classification or regression tasks on clinical data.