irg

irg performs Granger-causal analysis on irregularly sampled time series to identify causal relationships in dynamic transcriptome and intercellular signaling data.

Key Features:

• Statistical framework: Introduces a statistical framework for identifying Granger-causal relationships within and between multivariate time series characterized by irregular sampling intervals.
• Adaptation to irregular sampling: Incorporates engineering domain functions that adapt the model's dependence structure according to specific sampling times to handle unevenly spaced data.
• Parameter estimation and testing: Uses maximum-likelihood estimation for bivariate time series parameter estimation and employs bootstrap procedures for small samples or asymptotic methods for larger datasets for hypothesis testing of Granger-causality.
• Application to Arabidopsis thaliana data: Applied to nitrogen-response data in Arabidopsis thaliana, identifying 3078 significant gene–gene interactions and linking many causal genes to known and predicted mobile transcripts involved in long-distance nitrogen signaling.
• System-wide causal relationship identification: Enables detection of coordinated biological processes and potential intercellular communication pathways across tissues or organs from sparse time series data.

Scientific Applications:

• Dynamic transcriptome analysis: Infers causal relationships from sparsely and unevenly sampled gene expression time series.
• Intercellular and long-distance signaling studies: Identifies candidate causal genes and mobile transcripts involved in long-distance nitrogen signaling.
• Gene regulatory and signaling network inference: Reconstructs gene–gene interaction networks from irregularly sampled biological time series.

Methodology:

The approach adapts Granger-causal analysis to irregular sampling by adjusting the model dependence structure according to sampling times, estimating parameters by maximum-likelihood on bivariate time series, and using bootstrap procedures for small samples or asymptotic tests for larger datasets.