High-dimensional omics data are often generated in studies with limited sample sizes, posing major challenges for model-based clustering methods such as Gaussian mixture models, which may become unstable and generalize poorly in the presence of complex mixture structure. To address this, we developed Praxis-BGM, a natural-gradient variational inference framework for Gaussian mixture models that supports semi-supervised transfer learning through informative priors derived from large-scale reference datasets with well-defined cluster structure. These priors can encode cluster-specific means, covariance matrices, and structural connectivity, enabling robust knowledge transfer from a source domain to improve clustering in a smaller target dataset. This folder contains the latest Praxis implementation in the installable Praxis-BGM repository format, along with application examples.