Similarity

• Jaccard index
• Cosine similarity
• Hamming distance
• Connection Specificity Index
• Network similarity
• Cluster similarity, Cluster validation, Evaluation of community detection methods

Embedding space

How to measure similarity between embedding vectors obtained by Neural network models like word2vec, LLMs, Convolutional neural network, or Graph neural network?

Let’s think about two vectors $\mathbf{u}$ and $\mathbf{v}$. The most intuitive “distance” is the Euclidean distance: $$d_e (\mathbf{u}, \mathbf{v}) = \sqrt{ \sum_i (u_i - v_i)^2 }$$ Cosine distance

vs. Angular distance: Cosine similarity vs angular distance

References

• Nature methods: Using networks to measure similarity between genes: association index selection
  • Connection Specificity Index - A High-Resolution C. elegans Essential Gene Network Based on Phenotypic Profiling of a Complex Tissue