Automatic mutual information noise omission (AMINO)

Introduction

AMINO is a method that can be used to cluster a large set of Order Parameters (OPs) and then obtain a representative one from each of the clusters. The aim is to remove non-significant OPs and reduce their redundancy, in order to efficiently select a Reaction Coordinate (RC) by employing the representative OPs found as an input for other methods (e.g. PCA, TICA, RAVE).

The required input consists in a large set of OPs obtained from either a short unbiased simulation or from a biased one appropriately reweighted. Then AMINO steps can be summarized as follows:

  1. OPs are clustered

  2. A representative OP is determined for each cluster

  3. The appropriate number of clusters is determined

OPs are clustered using a Mutual Information based distance matrix with a modified KMeans clustering approach employing a dissimilarity matrix for centroid initialization. Finally, the proper number of clusters is determined using rate distortion theory as the driving principle.

https://pubs.rsc.org/en/content/articlelanding/2020/me/c9me00115h

 

Example Cases

Automatic mutual information noise omission (AMINO) was used in the following examples: