| Re: How are compounds clustered? [message #178 is a reply to message #177] |
Fri, 13 May 2016 21:22  |
thomas
Messages: 715
Registered: June 2014
|
Senior Member |
|
|
You are right. It is an agglomerative hierarchical clustering. It starts by calculating the complete similarity matrix between all molecules. Therefore it can be used by any descriptor, not just vector based ones. At the beginning every molecule already represents a cluster. Stepwise always those two clusters are merged, which are the most similar ones. The respective two rows and columns of the similarity matrix are then also merged by simply calculating a weighted mean between any two similarity values. The merging continues until the stop criterion is met.
Thomas
|
|
|
|
Search
Help
Members
Register
Login
Home
