Clustering Data

Abstract

Clustering reduces the size of the data by replacing individual genes with artificial super-genes that can be treated as single nodes for the purposes of network inference. By clustering genes that work together as a preprocessing step, we can improve the accuracy of the resulting network by reducing variance due to noise on individual genes. The goal is to generate clusters while losing the minimum amount of information in the dataset (and perhaps even make certain relationships stronger!). For example, if there are two genes that both behave in exactly the same way across the experimental conditions of interest, then little to no information is lost if you treat them as though they were a single “gene”.