Gene regulatory networks describes the transcriptional relationships between regulators (e.g., Transcription factors) and their target genes. Identifying gene regulatory networks is important to understand biological systems. Moreover, diseases such as cancers arises from dys-regulated gene regulation program. Thus, gene regulatory network identification is also crucial for elucidating mechanism of various diseases. In this dissertation, I introduce a series of studies that address three major challenges in gene regulatory network inference problems: 1) capturing dynamic properties of gene regulation process 2) identifying context-specific gene regulatory networks 3) transcription factor binding site prediction. In the first study, I introduce BTNET, a boosted tree based method that infers gene regulatory networks from time-series gene expression data. BTNET was designed to capture dynamic properties of gene regulation using time-series gene expression data. BTNET quantitatively outperformed the previous state-of-the-art methods and it was also validated with biological experiments. In the second study, I introduce CONFIGURE, a pipeline that identifies context specific regulatory modules. Gene regulatory networks are context (i.e., phenotype) specific. CONFIGURE was designed to identify regulatory modules representing each context using machine learning and statistical methods. CONFIGURE was validated in the context of breast cancer. CONFIGURE was shown to accurately classify the subtypes of breast cancer and also extract relevant regulatory modules of basal-like type breast cancer. In the last study, I introduce TBiNet, an attention based deep neural network for predicting transcription factor binding sites. Transcription factor (TF)-binding motifs contain strong signal of transcription factor binding sites. By adopting attention mechanism, TBiNet was design to assign higher weights on the TF-binding motifs when predicting transcription factor binding sites. TBiNet significantly outperformed the previous state-of-the-art methods on the ENCODE ChIP-seq dataset. Moreover, TBiNet was shown to extract TF-binding motifs more effectively than the previous methods.