In late 1994, BLAST was fast, but it wasn't as sensitive as programs that produced gapped alignments, such as Bill Pearson's FASTA. To keep the statistics in sync with the new search algorithm, I conjectured that the statistics of Samuel Karlin and Stephen Altschul might be empirically applied to the interpretation of gapped alignment scores.
By early 1995, Altschul at the NCBI [National Center for Biotechnology Information] had obtained results supporting this approach. Later, I evaluated parameters for the nascent WU BLASTP 2.0, using a test recently published by Bill Pearson.1 The program ran only about 10% slower with gapped alignments but exhibited sensitivity (shown here) that was essentially the same as FASTA and Smith-Waterman. A complete gapped BLAST package was released in May of 1996.2
--Warren Gish, Washington University in St. Louis