Originally posted as a Reader Response on 5th February, 2007

I'd like to add a distinction to the well-written article by Fran Lewitter on "Moving Education Forward."

I refer to the remark:
"Recently, I asked an MIT Biology Professor if his newer graduate students were more knowledgeable in bioinformatics and computational biology than in past years. [...] They may also have used computational tools on the Web but with little understanding of the guts of the algorithm."

Of course, BIOINFORMATICS students should "understand the guts of the algorithm." The question, however, was about BIOLOGY students. I advocate that education of biologists should be model-oriented. Algorithms are being improved continuously, (good) models last. I suggest the biologists should be aware that there is a model underlying each tool and each algorithm. Students should learn about successful models -- in phylogenetics, sequence comparison, structure family modeling, and more. With respect to algorithms, they should be familiar with three facts:

(1) There is a model behind each algorithm. The model relates to biology and determines whether a computational approach is appropriate to the biological question at hand. Each model has intrinsic limitations, which should be known. Algorithms are grey boxes that solve models, with only two abstract properties that are relevant.

(2) Some algorithms return a complete (optimal) answer with respect to their model; some only approximate such a solution. The notion of exact versus heuristic algorithms should be clear. Algorithmic limitations or inadequacies should be distinguished from model limitations.

(3) The notion of algorithmic complexity - alias asymptotic runtime and space requirements - should be familiar to biologists. It determines which tool based on which algorithm is practical for the given data size.

Good judgment in these three respects allows us to choose the best available tool for the problem at hand. And also, it should enable biologists to give bioinformaticians competent feedback on potential shortcomings of tools. Let's have the bioinformaticians worry about algorithm improvements in the right place.

Kay Nieselt (Tuebingen) and myself have recently conducted an "Advanced Course on Computational RNA Biology" for biologists,
strictly modeled after these principles. The response has been so positive that we plan to repeat it on a larger scale.

Submitted by: Robert Giegerich
E-mail: robert@techfak.uni-bielefeld.de
Occupation: Professor
Bielefeld University, Germany