CD-HIT is an ultra-fast program for the clustering of biological sequences. This paper describes significant improvements and new developments of CD-HIT. I see three primary reasons for its being highly cited. 1) There is a great need for a fast sequence clustering program to help various sequence-based research, and there are not many such programs available. CD-HIT may be the only one that can handle a huge dataset of tens of millions of sequences. 2) A previous version of CD-HIT exists which has attracted many users and citations. 3) The CD-HIT program is reasonably well maintained, distributed, and documented, so that a user can easily download and apply it without experiencing technical problems.

“This paper describes a computer program and the underlying algorithm for rapid sequence clustering and comparing.”

This paper describes a computer program and the underlying algorithm for rapid sequence clustering and comparing. Some earlier works had previously been published—see Bioinformatics 17:282, (2001), and Bioinformatics 18: 77, (2002)—but several significant improvements were first described in this paper.

As a result of high-throughput genome sequencing projects, researchers are facing serious challenges and problems from the explosive growth of public sequence databases. Routine sequence analysis is getting more computationally expensive and more complex. Also, the general growth of databases is quite uneven, which may lead to biased conclusions.

An efficient clustering method is the key to addressing these challenges and overcoming the problems. However, sequence clustering is also quite computationally intensive. Our contribution here is an algorithm which could speed up the clustering calculation by two to three orders of magnitude. So, a user can easily apply our method to his/her sequences, even with a very huge dataset.

I have witnessed the increasing growth of public sequence databases since I first began working in the bioinformatics field 10 years ago. In our research, in order to reduce the effort of sequence analysis, we clustered the sequence datasets and used only its representatives. We noticed that such usage actually improved the result, and similar findings were reported in the literature.

With the database continuing to grow, sequence clustering itself became a challenge: this is why I began a search for an efficient algorithm for rapid sequence clustering and to concentrate on writing the CD-HIT program. The development has been fairly smooth, and I’ve received considerable support, help, and suggestions from many CD-HIT users. The only problem has been the dearth of grant support—much of the program development was completed in my spare time.

Sequence clustering methods will have numerous applications in various fields for as long as a large amount of biological sequences are being analyzed. Among the unique features of our CD-HIT program are its ultra-high speed and its inherent capability of handling a huge amount of sequences. I envision several new and important applications, such as the analysis of metagenomic sequences. I can also foresee potential developments which will further improve the clustering method, and which will simultaneously become invaluable throughout the worldwide research community.