We introduce GAMLC (Graphical models for Mixed-type data: Latent Gaussianization with Covariates), a unified framework for estimating covariate-dependent graphical models with mixed response types (categorical, ordinal, and continuous). At its core, GAMLC employs a covariate-dependent latent Gaussian graphical model to enable individualized network estimation that flexibly adapts to subject-specific characteristics. To address high-dimensional settings, we develop a scalable penalized expectation-maximization (PEM) algorithm with principled regularization. Theoretically, we establish novel guarantees for graph recovery and parameter estimation, including the first minimax lower bound for latent parameters in this setting. Extensive simulations demonstrate that GAMLC consistently outperforms state-of-the-art methods across diverse scenarios. An application to the National Health and Nutrition Examination Survey (NHANES) data reveals interpretable multimorbidity networks and significantly improves disease outcome prediction. By offering a unified, theoretically grounded solution for heterogeneous, covariate-dependent data, GAMLC advances network estimation in biomedical and social science research.