Graph-Based Substitute Discovery

This project uses a heterogeneous graph representation of raw materials and molecular interactions to identify compatible substitutes in cosmetic formulations. Metapath-based embeddings capture structural and functional similarity across ingredients.

Key Features

• Heterogeneous graph modeling of molecules and interactions.
• Metapath2vec embeddings for similarity search.
• Candidate ranking for formulation compatibility.

Technologies Used

• Python, DGL, Gensim, PyG