Their previous offering (S.F. Altschul, et al., "Basic local alignment search tool," J. Mol. Biol., 215:403-10, 1990) is the most highly cited paper published in this decade, now with more than 7,800 citations to its credit. That is because the tool is, quite simply, indispensable. Any researcher who has uncovered the sequence of a stretch of DNA or a protein wants to know what the sequence does. The easiest way to find out is to see whether it resembles another sequence that somebody else has already investigated. And the easiest way to do that is to submit the sequence to the NCBI's suite of computer programs and ask BLAST to hunt for matches among the databases of stored sequences.

   As sequencing has come to occupy more and more research effort, it is not surprising that BLAST came to be used (and cited) more and more, but partly for unforeseen reasons. The programs were designed to identify strong sequence similarities swiftly, as Stephen Altschul tells Science Watch, but "it was something of a surprise that the BLAST programs proved sensitive enough to weak similarities that they could be used for general-purpose database searches." That sensitivity, coupled with their speed and, perhaps most importantly, the ability to incorporate rigorous statistical measures to help researchers assess the significance of sequence alignments, is what gave BLAST the edge over other similar programs.

   Altschul and his colleagues have now modified the programs to make them more sensitive to weak similarities while at the same time making them work three times faster. This has been achieved in two main ways. Firstly, BLAST is now able to span gaps in either the query sequence or the target. Other programs could do that before, but BLAST does it faster. Perhaps more important, a new kind of program, PSI-BLAST, can look for higher-order patterns in a protein sequence, often called a protein profile. This enables researchers to widen their net and snag evolutionarily distant relationships that a traditional search program might miss.

   Profile analysis of this kind is not new, but it used to require several different programs and took a long time to run. "PSI-BLAST can perhaps be viewed as the Model T," Altschul tells Science Watch. "It was by no means the first, but it makes readily accessible a powerful tool that previously was difficult and time-consuming to use, even for experts."

   The sensitivity of new pattern-matching programs is revealing unexpected truths about evolution. Instead of purely vertical descent of DNA from one generation to the next, with occasional splits for speciation, the ability to pick out distant relationships is revealing very large scale horizontal transfer, especially among bacteria. This "should be considered a major evolutionary mode rather than some sort of an exception," says Eugene Koonin, also at NCBI, who has been using PSI-BLAST to study prokaryote evolution. "I expect that this will indeed make the work of molecular phylogeneticists harder," adds Altschul, "but only because the truth is more complicated than had been expected."

   The speed issue is a difficult one to grasp. At the moment, to search a typical protein of 300 amino acids against a database of 100,000,000 amino acids takes about 30 seconds. A DNA sequence of 10,000 bases takes slightly longer, 45 seconds. But as Altschul points out, "practically speaking, a three-fold increase in speed probably translates mainly into fewer computers at NCBI, and therefore saved tax dollars."