On this page
Today a biologist who finds a new protein can compare it with huge public collections of known sequences in minutes. Someone had to build the first collection. That was largely the work of Margaret Dayhoff, a chemist who saw early that computers could store, compare and make sense of the protein sequences that biochemists were starting to publish. This post covers who she was, what was in her Atlas, and how a sequence comparison works.
Who was Margaret Dayhoff?
Her obituary in the Journal of Molecular Evolution gives her dates: 11 March 1925 to 5 February 1983. Historian Bruno Strasser's study in the Journal of the History of Biology fills in her career. After a BA in mathematics and an MA in chemistry, she earned a PhD in quantum chemistry in 1948 under George Kimball at Columbia University. As a fellow at the Watson IBM Computing Laboratory in 1947-1948 she used punch card machines to calculate resonance energies of small molecules.
She worked at the Rockefeller Institute and the University of Maryland, then joined the National Biomedical Research Foundation (NBRF) in 1960. Robert Ledley had founded the NBRF that year to explore how electronic computers could help biomedical research. Dayhoff later became professor of physiology and biophysics at Georgetown University and president of the Biophysical Society (1980-1981).
Why did protein sequences need collecting?
Frederick Sanger had determined the first complete protein sequence, insulin, by 1953 (see our post on Fred Sanger and sequencing). Through the 1950s and early 1960s, more sequences appeared, scattered across journals. Biochemists compared the same protein from different species to find which positions mattered.
Dayhoff put the case in a 1967 letter that Strasser quotes: "there is a tremendous amount of information regarding evolutionary history and biochemical function implicit in each sequence and the number of known sequences is growing explosively."
What was in the 1965 Atlas?
The first edition of the Atlas of Protein Sequence and Structure, by Dayhoff, Richard Eck and two NBRF colleagues, was just under 100 pages and held the sequences of around 70 proteins, mainly cytochromes c, haemoglobins and fibrinopeptides from various species. Strasser describes each page: the protein and organism ("Hemoglobin beta - gorilla", for example), the sequence in both three-letter and one-letter amino acid codes, the amino acid composition, notes on how the data were obtained, and a reference to the source. The Atlas also aligned the same protein from several organisms, so a reader could see at a glance which regions were conserved.
Behind the book was a computer. Every sequence was entered on punch cards, one card per entry. Dayhoff and Eck adopted a one-letter notation for amino acids to save computer memory and to make alignments readable on fixed-width printers.
The authors asked readers to send in new sequences and corrections. The Protein Information Resource records that the Atlas appeared from 1965 to 1978 under her editorship, that her group pioneered computer methods for comparing sequences and inferring evolutionary histories, and that it introduced the protein superfamily concept in 1975.
What did the comparisons reveal?
Using a program that matched the sequence of ferredoxin against itself, Eck and Dayhoff argued in Science in 1966 that the protein had evolved by duplication of a much shorter ancestral sequence. Strasser notes that two other groups reached the same conclusion that year.
Another program, ALLELE, used all the data in the Atlas to build a matrix of probabilities of one amino acid replacing another. That matters because two sequences can differ at a position after several changes, not just one. In Strasser's example, two cytochrome c sequences with 52 visible differences were estimated to have undergone as many as 92 changes.
Worked example: comparing two sequences
Take two short sample sequences in one-letter code (made up for practice, not real proteins):
Sequence A: P L V A S T R E N G
Sequence B: P L I A S T R D N G
Line them up position by position. Both have 10 positions.
Count mismatches: position 3 (V vs I) and position 8 (E vs D). That is 2 differences.
Identity = matching positions ÷ total = 8 ÷ 10 = 80%.
A common mistake is to treat 2 differences as exactly 2 mutations. As the cytochrome c example shows, a position can change more than once, or change and change back, so the true number of changes is usually higher than the visible count. Another mistake is comparing sequences without aligning them first; one missing letter shifts every position after it.
If the four levels of protein structure are still hazy, our guide to protein structure levels explains how the sequence is the first of them. For worked problems on topics like this, the Fermi Problem Library has a step-by-step solver that explains the reasoning rather than printing an answer.
What can a student learn from how Dayhoff worked?
Dayhoff did not sequence proteins herself. She saw that many people's results, collected in one consistent format, could answer questions none of them could answer alone. She chose a notation that made the data easy to compare, and she built tools to analyse it. After her death in 1983, the PIR notes, Winona Barker and Robert Ledley took over the project, and the NBRF established the Protein Information Resource in 1984.
Try it now: write two eight-letter sample sequences that differ at three positions, then calculate their percentage identity.
