In the quest to understand large-scale collective behavior in active matter, the complexity of hydrodynamic and phoretic interactions remains a fundamental challenge. To date, most works either focus on minimal models that do not (fully) account for these interactions, or explore relatively small systems. The present work develops a generic method that combines high-fidelity simulations with symmetry-preserving descriptors and neural networks to predict hydro-phoretic interactions directly from particle coordinates (effective interactions). This method enables, for the first time, self-contained particle-only simulations and theories with full hydro-phoretic pair interactions. Here, this dataset contains the data and code associated with the research project “Learning Hydro-Phoretic Interactions in Active Matter”. It includes raw figure data, scripts for generating the figures, and training/testing datasets for machine learning models that predict velocity and angular velocity in active colloidal systems. The dataset supports a machine learning-based framework for learning hydro-phoretic interactions from high-fidelity simulations.