Why do you think your paper is highly cited?

“Clustering” is a commonly used exploratory data analysis technique for attempting to discover patterns in large, complex data sets such as gene expression microarray experiments.  The explosion of such data in recent years has spawned invention of new clustering algorithms and renewed interest in old ones.  Which of these many methods are best for particular tasks or particular kinds of data?  It's easy enough to compare clustering algorithms on data sets where the "ideal clustering" is known in advance, but these cases are usually toy examples.  Our paper provides one of the few methods available for comparative evaluation of clustering algorithms on real data where the "right answer" isn't known in advance.

  Does it describe a new discovery or a new methodology that's useful to others?

It's a new methodology that's basically applicable for the comparison of any clustering algorithms on any particular data set.

  What were some of the circumstances that led you to do this research?

In the early work on microarrays, every research group seemed to have their own favorite clustering method, and all clustering algorithms find "clusters"—that's their job.  But how could you tell whether those clusters were good ones?  We were looking for a more systematic, data-driven way to evaluate the methods, so that researchers could have more confidence in their results.

  Could you summarize the significance of your paper in layman's terms?

"Clustering" is the process of dividing data points into groups or clusters so that, hopefully, the points in each group are more similar to each other than to points in other groups.  With luck, this will lead you to some useful hypotheses about the biological system you're studying, e.g., the genes in cluster A are related to such-and-such a function, while those in cluster B serve a different purpose.  Unfortunately, these divisions usually aren't clear-cut, and different clustering algorithms make different choices.  We proposed a way to test the "quality" of the algorithms without knowing the best clustering in advance.