Regulation of transcription factor activity is dynamically changed across cellular conditions and disease subtypes. The identification of biological modulators contributing to context-specific gene regulation is one of the challenging tasks in systems biology, in order to understand and control cellular responses across different genetic backgrounds and environmental conditions. Previous approaches for the identification of biological modulators from gene expression data are restricted to the capturing of a particular type of a three-way dependency between a regulator, its target gene, and a modulator, and these methods cannot describe complex regulation structure, such as where multiple regulators, their target genes, and modulators are functionally related. Here, we propose a statistical method for the identification of biological modulators by capturing multivariate local dependencies, based on energy statistics, which is a class of statistics based on distances. Subsequently, out method assigns a measure of statistical significance to each candidate modulator by a permutation test. We compared our approach with a leading competitor for the identification of modulators, and illustrated its performance both through the simulation and real data analysis. GIMLET is implemented with R (≥ 3.2.2) and is available from github (https://github.com/tshimam/GIMLET).