Søren Brunak is a Professor of Disease Systems Biology at the University of Copenhagen and Professor of Bioinformatics at the Technical University of Denmark. He is also Research Director at the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen Medical School.
He sent the following comment:
I agree that access to data was key to AlphaFold and to previous work in this field. I still have in my office the original magnetic tapes that contained the PDB data used to train the machine learning methods in the 1988 and 1990 papers:
The AlphaFold method predicts inter-residue distance distributions and then converts predicted distance probabilities into statistical potential for energy minimization to obtain 3D coordinates.
In 1990, we were first to predict the distance matrix for proteins by neural networks (Bohr, 1990). At my Center for Biological Sequence Analysis, we later developed energy minimization methods that converted the distance matrices into coordinates. Many other people did that, but the 1990 paper made the important step of designing neural networks to predict distance matrices. We also organized a meeting in 1993 around distance-based methods and published the contributions as proceedings (Eds. Bohr & Brunak, 1994)
Bohr H, Bohr J, Brunak S, Cotterill RM, Fredholm H, Lautrup B, Petersen SB. A novel approach to prediction of the 3-dimensional structures of protein backbones by neural networks. FEBS Lett. 1990 Feb 12;261(1):43-6. doi: 10.1016/0014-5793(90)80632-s. PMID: 19928342.
Eds. Bohr H, Brunak S. Protein Structure by Distance Analysis. IOS Press, Amsterdam, 352 pp., 1994.