mims-harvard/scikit-fusion

Implements collective matrix factorization across heterogeneous, multi-relational datasets by jointly decomposing multiple interconnected matrices while sharing latent factors across object types. Supports both learning on complete fusion graphs and transforming new data into learned latent spaces. Built on NumPy/SciPy with algorithms optimized for large-scale inference across biological networks, drug-disease relationships, and gene annotation tasks.

151 stars. No commits in the last 6 months.