JMIR Publications

ABSTRACT

Background:

Over the last few decades, numerous amounts of biological data have been generated and urged scientists to find the correlation among genes responsible for different diseases. The clustering technique represents such correlation among multiple species and genes. So, it has become of immense importance to find a proper distance-linkage metric to create clusters from variant biological datasets. Pleiotropy also plays a vital role for a gene to be expressed differently and cause different outcomes in living beings. It has become an important research challenge to find the pleiotropy of genes responsible for different diseases.

Objective:

Our research aims to find the best distance–linkage method to create accurate clusters from variant datasets and identify the common genes that responsible for different cancers.

Methods:

We have used three distance metrics - Euclidean, Maximum, and Manhattan distance, and four linkage methods – Single, Complete, Average, and Ward. We have proposed a fitness function combining silhouette width and within-cluster distance for comparing the quality between different sets of clusters

Results:

Our study finds that the maximum distance metric gives the best quality of clusters. Moreover, for medium dataset average linkage and large dataset ward linkage method works best. Ensemble clustering doesn’t improve the result. We have also identified genes that are responsible for three different cancer and validated our findings by gene enrichment

Conclusions:

Accuracy is very important in clustering and in our research we have measured the accuracy of different clustering techniques. Other research field also can find inspiration from this paper if their dataset resembles ours.