Moving Beyond Local Patterns: New Ways to Reason about Biological Network Comparisons
summary
Method embeds each network by measuring diffusion-based distances to a selected set of landmarks (usually one-to-one homologs) and compares nodes in a 200–500 dimensional space so structure alone drives alignment. Uses diffusion-based landmark embedding with metrics like DSD and heat kernels, reduces comparisons with cosine or L1 norms and performs Hungarian matching for one-to-one outputs. Tests on PPI networks (human ~14,000, yeast ~5,000) show robustness to thinning/rewiring up to large perturbations, recovers withheld homologs, and flags known MMS repair and breast cancer coexpression hits; estimated metric dimension ~800.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 0:22General Biological Questions
- 0:49Old School Comparisons
- 2:44Diffusion
- 3:47Pioneer Plaque
- 4:27New Representation
- 5:14Metric Dimension
- 7:45Making Comparisons
- 8:44Preserved Work
- 8:57Network Alignment
- 9:38Alignment
- 11:11Diffusion Distance
- 12:32Clustering
- 12:54Matching
- 14:30Preliminary Results
- 15:41Ground Truth
- 16:41Thin
- 17:25Scramble
- 19:27Validation Techniques
- 20:39CrossValidation
- 21:36Resnick Score
- 23:06Comparisons
- 27:18Authors
- 27:58Cosine