1 This prediction was written in October 2023, and we have not tried to update it since. It would in any case be difficult given extremely rapid developments.

2 Lately, increasing numbers of protein structures have been determined using cryo-electron microscopy, a process for which three scientists – Jacques Dubochet, Joachim Frank and Richard Henderson – received the Nobel Prize in chemistry in 2017.

3 Hall, Stephen S., ‘Revolution Postponed: Why the Human Genome Project has been Disappointing,’ Scientific American, 1 October 2010.

4 Fridovich-Keil, J. L. https://www.britannica.com/event/Human-Genome-Project/Advances-based-on-the-HGP, last updated August 22, 2023.

5 Caulfield, Timothy. https://www.ualberta.ca/folio/2019/12/commentary--the-problem-with-personalized-health-information.html. December 16, 2019.

6 Not coincidentally, the philosopher G. W. Leibniz, who inspired Gilles Deleuze’s ‘The Fold’, found support for his metaphysics in the observations of van Leewenhoek, with whom he maintained a four decade-long correspondence. ‘Now nothing better corroborates the incomparable wisdom of God,’ he wrote in the Monadologie, ‘than the structure of the works of nature…which appears when we study them more closely with a microscope…A man in Delft (van Leeuwenhoek) has accomplished wonders at it.’ Deleuze (1993: 31) describes Leibniz’ baroque metaphysics as consisting of folds and pleats, which we can ‘behold…in the microscope, collectivities through the folds of dust.’

7 Among the investors were Peter Thiel and Elon Musk, and the ‘long-termist’ philosopher Nick Bostrom was hired for DeepMind’s ‘Ethics and Society’ unit. For this discussion, we leave these ultra-libertarian, right-wing underpinnings to the side.

8 Since the early 1970s, around 194,000 structures have been determined experimentally and deposited in the Protein Data Bank (PDB). They constitute the information from which Alphafold ‘learned’ to predict protein structure.

9 The competition uses protein sequences for which experimental 3D structures have recently been determined but are not yet publicly accessible.

10 At https://alphafold.ebi.ac.uk.

11 Editorial. ‘Artificial Intelligence in Structural Biology Is Here to Stay,’ Nature 595, 625–26 (2021) https://www.nature.com/articles/d41586-021-02037-0.

12 At https://www.deepmind.com/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology.

13 The co-evolution of amino acids in protein interfaces likely enables the original Alphafold2 to model protein complexes in some cases. Alphafold multimer (Evans et al 2022), which was trained specifically on protein complexes, shows higher accuracy. One of us has recently used these deep learning models to predict the structure of a 15-subunit complex (Petriman et al. 2022).

14 Lowe, Derek. https://www.chemistryworld.com/opinion/why-alphafold-wont-revolutionise-drug-discovery/4016051.article. 5 August 2022.