(374i) Identifying Causality and Estimating Time Delay for Non-Stationary Bivariate Time Series Using Novel Time Delay Rényi Symbolic Transfer Entropy

Identifying causality and estimating time delay in biological time-series data is challenging due to non-stationarity, non-linear relationships, and bidirectional couplings within the system. To address this, we developed a model-free information-theoretic measure called Time Delay Rényi Symbolic Transfer Entropy (TDRSTE) and applied it to infer causality and estimate time delay among variables using high-throughput omics time series data. TDRSTE effectively combines the generalization feature of Rényi Transfer Entropy (RTE) with the applicability of Symbolic Transfer Entropy (STE) to non-stationary data, offering fast computation through symbolic analysis. We applied the TDRSTE framework to multi-omics data from mouse bone marrow derived macrophages. We validated the efficacy of TDRSTE by accurately estimating causal time delays for synthetic datasets as well as a real-world dataset on eicosanoids, which are part of Arachidonic acid (AA) metabolic network. Our study integrates multi-omics data and predicts causality along with the time delays for bivariate time series of AA and cytokines. The TDRSTE framework captured 25 out of 31 reaction connections in the AA network, attaining an excellent prediction accuracy (true positive rate) of 80.6%. Our results indicate a potential causal link between AA and cytokines, where AA may initiate the secretion of cytokines like TNFα, IL1α, IL18, and IL10. Conversely, cytokines such as IL6 and IL1β may have an early causal impact on AA. These results pave the way for further exploration with suitably designed experiments to quantify causal influence in future investigations.