Clustering MD frames

Given thousands of MD snapshots, how do we divide them into clusters and pick representative snapshots for each cluster?

From the clustering, we need to know:

• the centroids
• weight

Clustering at a glance

Multi-dimensional clustering

Defining distance

Distance based on RMSD of aligned structures

Distance based on Structure signature

Comparing two distance measurement

Clustering frames based on distance

One dimensional clustering

Pros:

• Fast
• Intuitive
• Can visually identify number of clusters

Cons:

• Use partial information
• Less accurate, may miss-place elements

Find a numerical value to represent a structure

Clustering in one dimension

Enhance the accuracy by dual clustering

Compare all results

Accuracy

Efficiency