OIPCQ

OIPCQ infers gene regulatory network (GRN) structure using quantile-based thresholds for Conditional Mutual Information and extended path-based conditional gene selection to produce order-independent network reconstructions.

Key Features:

• Order Independence: Eliminates dependency on gene ordering by employing quantile values to set thresholds for Conditional Mutual Information (CMI) tests, producing consistent network structures across gene permutations.
• Quantile-Based Thresholds: Uses quantile-based thresholds for independence testing of CMI to dynamically adjust significance thresholds and improve causal inference accuracy.
• Extended Path Consideration: Evaluates paths of length greater than or equal to 2 (rather than only length 2) for conditional gene selection, enhancing selection of conditional genes and improving performance and computational efficiency.
• Application in GRN Inference: Applied to in silico networks from the DREAM3 and DREAM4 challenges and to real-world GRNs including the SOS DNA network from Escherichia coli and an acute myeloid leukemia GRN reconstructed from RNA sequencing data in The Cancer Genome Atlas (TCGA).
• Scientific Contributions: In studies of acute myeloid leukemia, identified regulators BCLAF1 and NRSF and highlighted the relevance of ZBTB7A and PU1 in leukemia biology (reported by Zhang et al.).

Scientific Applications:

• DREAM benchmarking: Benchmarking and validation on DREAM3 and DREAM4 in silico network challenges.
• Bacterial GRN reconstruction: Reconstruction of the SOS DNA regulatory network from Escherichia coli.
• Cancer GRN reconstruction: Reconstruction of an acute myeloid leukemia gene regulatory network using RNA sequencing data from TCGA.
• Regulator discovery: Identification of candidate regulators such as BCLAF1, NRSF, ZBTB7A, and PU1 in leukemia studies.

Methodology:

Performs Conditional Mutual Information (CMI) tests (functions such as cmi and Compare; OIPCQ2 extends this with MI2), applies quantile-based thresholds for independence testing, and enhances the Path Consistency algorithm by considering conditional gene sets derived from paths of length ≥2.