Basics of Phylogenetic-like Trees
Preprocess:
- Describe compounds with (MACCS_like) keys
- Keys used are actually Smarts queries
Process:
- Cluster active compounds with SOM (smiles strings)
- Detect “hotspots” in the SOM
- Self Organizing Map - Kohonen clustering algorithm
- Stable, robust, unsupervised neural network
- Locally organizes inputs into 2D neighborhoods
- “Hotspots” - neighborhoods of clusters that contain molecules similar with respect to some axis
- Hotspots determined by set of expert system rules
Structural Family Identification