Nobel laureates Alan Hodgkin and Andrew Huxley are sometimes called the fathers of computational neuroscience, because they developed the first mathematical theory to explain how neurons communicate with each other with electrical signals. I read in a textbook that when they published their model in 1952, all the numerical calculations were done manually, because the computer at Cambridge was not working. It was “computational” all right, but it was the labour intensive variety. I can see that scientists quickly realized that they needed some help to speed up the computation, but I hadn’t wondered who it was who did the first Hodgkin-Huxley simulation on a computer.
I became intrigued by the question, because in a recent paper about dendritic computation, I encountered a reference to a 1973 paper by Dodge and Cooley, which was oddly published in a journal called the IBM Journal of Research and Development (a peer-reviewed journal that is still published by IBM today). Why IBM? I never thought that a technology company was involved in the early development of computational neuroscience, but then it occurred to me that maybe in those days, IBM was one of the few places in the world which had the computational power to simulate a neuron.
Luckily, I found a paper that gives some context to this history. The first computer simulation of the Hodgkin-Huxley equations was published in 1955 by Kenneth Cole and his colleagues at the National Institute of Health. The computation was performed on SEAC - a vacuum tube computer built by the National Bureau of Standards. It took SEAC 30 minutes to calculate one action potential, which is really not bad at all. I would have guessed days.
In the next couple of years, a couple of attempts were made to numerically solve the Hodgkin-Huxley equations. A notable example was published by Fitzbugh and Antosiewicz in 1959, this time with the help of an IBM 704 mainframe. It was still vacuum tube-based, but it was a commercial product with instructions to operate on floating point numbers. The arrival of transistor computers, and the high-level programming language Fortran, finally made the simulation of biological systems more practical. IBM’s own research department became interested in this problem, and published its first computational neuroscience paper in 1966, which was the first to simulate the propagation of action potentials on axons. The wikipedia’s entry on the IBM 7094 gives us some idea about how expensive this simulation was: the price for renting the computer for a month was about half a million in today’s money.
The authors of the paper, both IBM scientists, were interesting and give some insights about the world of scientific computation of the 60’s. James Cooley was an electrical engineer who invented the Cooley–Tukey algorithm for Fast Fourier Transform. Fred Dodge was associated with the lab of Nobel laureate Haldan Hartline, and helped to derive the transfer function of the eyes of the horseshoe crab. In other words, he was also a pioneer of computational vision research. It’s really fascinating history.