Compensated Transfer Entropy as a Tool for Reliably Estimating Information Transfer in Physiological Time Series

摘要: We present a framework for the estimation of transfer entropy (TE) under conditions typical physiological system analysis, featuring short multivariate time series and presence instantaneous causality (IC). The is based on recognizing that TE can be interpreted as difference between two conditional (CE) terms, builds an efficient CE estimator compensates bias occurring high dimensional conditioning vectors follows sequential embedding procedure whereby are formed progressively according to criterion minimization. issue IC faced accounting zero-lag interactions alternative empirical strategies: if deemed physiologically meaningful, effects assimilated lagged make them causally relevant; not, incorporated in both terms obtain compensation. resulting compensated (cTE) tested simulated series, showing its utilization improves sensitivity (from 61% 96%) specificity 5/6 0/6 false positives) detection information respectively when effect meaningful non-meaningful. Then, it evaluated examples cardiovascular neurological supporting feasibility proposed investigation mechanisms.