3DeFDR: Identifying cell type-specific looping interactions with empirical false discovery rate guided thresholding

Abstract

The mammalian genome is connected into tens of thousands of long-range looping interactions critically linked to spatiotemporal gene expression regulation. An important unanswered question is to what extent looping interactions change across developmental models, genetic perturbations, drug treatments, and disease states. Although methods exist for calling loops in single biological conditions, there is a severe shortage of computational tools for rigorous assessment of cell type-specific looping interactions across multiple biological conditions. Here we present 3DeFDR, a simple and effective statistical tool for classifying dynamic looping interactions across biological conditions from Chromosome-Conformation-Capture-Carbon-Copy (5C) data. 3DeFDR parses chromatin contacts into invariant and cell type-specific classes by thresholding on differences in modeled interaction strength signal across two or three cellular states. Thresholds are iteratively adjusted based on a target empirical false discovery rate computed between real and simulated 5C maps. 3DeFDR enables the sensitive detection of high-confidence looping interactions and markedly reduces false positives when benchmarked against a classic analysis of variance (ANOVA) test, our newly formulated parametric likelihood ratio test (3DLRT), and the leading Hi-C differential interaction caller diffHic. 3DeFDR also sensitively and specifically calls loops in Mb-scale genomic regions parsed from Hi-C data. Our work provides a statistical framework and an open-source coding library for identifying dynamic long-range looping interactions in high-resolution 5C data from multiple cellular conditions.