LINGER is a machine learning method that infers gene regulatory networks (GRNs) from single-cell multiome data by integrating atlas-scale bulk data and prior knowledge of transcription factor (TF)–regulatory element (RE) motifs. It achieves a fourfold to sevenfold increase in accuracy compared to existing methods, enabling enhanced interpretation of disease-associated variants and genes. LINGER uses lifelong learning to incorporate external bulk data, allowing the model to continuously update as new data become available. This approach addresses the challenge of limited single-cell data and complex model fitting. LINGER also enables the estimation of TF activity from gene expression data, identifying driver regulators. The method was validated using various datasets, including peripheral blood mononuclear cells (PBMCs), and demonstrated superior performance in predicting trans-regulatory and cis-regulatory strengths. LINGER was further used to identify key TFs and regulatory networks associated with diseases such as inflammatory bowel disease (IBD). It also identified driver regulators in acute myeloid leukemia (AML) and demonstrated the ability to predict gene expression changes in response to T cell receptor (TCR) stimulation. LINGER's integration of single-cell and bulk data, along with its lifelong learning approach, provides a robust framework for GRN inference and regulatory network analysis. The method's ability to incorporate prior knowledge and adapt to new data makes it a valuable tool for understanding complex gene regulatory mechanisms and their implications in disease.LINGER is a machine learning method that infers gene regulatory networks (GRNs) from single-cell multiome data by integrating atlas-scale bulk data and prior knowledge of transcription factor (TF)–regulatory element (RE) motifs. It achieves a fourfold to sevenfold increase in accuracy compared to existing methods, enabling enhanced interpretation of disease-associated variants and genes. LINGER uses lifelong learning to incorporate external bulk data, allowing the model to continuously update as new data become available. This approach addresses the challenge of limited single-cell data and complex model fitting. LINGER also enables the estimation of TF activity from gene expression data, identifying driver regulators. The method was validated using various datasets, including peripheral blood mononuclear cells (PBMCs), and demonstrated superior performance in predicting trans-regulatory and cis-regulatory strengths. LINGER was further used to identify key TFs and regulatory networks associated with diseases such as inflammatory bowel disease (IBD). It also identified driver regulators in acute myeloid leukemia (AML) and demonstrated the ability to predict gene expression changes in response to T cell receptor (TCR) stimulation. LINGER's integration of single-cell and bulk data, along with its lifelong learning approach, provides a robust framework for GRN inference and regulatory network analysis. The method's ability to incorporate prior knowledge and adapt to new data makes it a valuable tool for understanding complex gene regulatory mechanisms and their implications in disease.