As of two or three years ago, 170,000 proteins had been worked through by people.

The claim is that, roughly two or three years before July 2024 (about 2021–2022), the number of proteins worked through by human (experimental) effort was about 170,000. In context this almost certainly means protein structures determined experimentally, for example entries in the Protein Data Bank or the experimental structures AlphaFold was trained on. Judge against the documented count for 2021–2022, using the measure the speaker's context points to. True: a credible record (PDB statistics, or DeepMind/AlphaFold publications describing the training set or the experimental baseline) puts the figure for that period at about 170,000, roughly 150,000–190,000, since the claim is a round figure. Mostly true: the figure is right in order of magnitude and in substance, but the closest matching count falls a little outside that band (about 130,000–150,000 or 190,000–215,000), or it fits a year slightly outside the 2–3-year window. Mixed: the figure matches one reasonable reading (such as AlphaFold's training-set size) but clearly conflicts with another equally reasonable one (such as total PDB entries). Mostly false: the number is badly off (more than about 25% from every reasonable measure for the period), or it rests on a real figure misdescribed, for instance a count of protein entries or sequences presented as solved structures. False: records show the relevant count for 2021–2022 was of a different order of magnitude (for example tens of thousands or millions).

Set before the research began.

1. E1 Accelerating scientific discovery with AI Nobel Prize Outreach (Demis Hassabis, Google DeepMind, Nobel Lecture) · 8 December 2024

   “After decades of experimental work ~170,000 structures had been determined and collated in Protein Data Bank (PDB), an incredible resource that we used as a starting point to train AlphaFold”

2. E2 Annual Report 2021 RCSB Protein Data Bank

   “In 2021, 12,640 new PDB structures were added to the archive, for a total of 185,541 available entries as of January 1, 2022.”

3. E3 Protein Data Bank: A Comprehensive Review of 3D Structure Holdings and Worldwide Utilization by Researchers, Educators, and Students - PMC Biomolecules (MDPI), via PubMed Central; author Christine Zardecki, Rutgers · 4 October 2022

   “As of mid-2022, the archive contained 166,894 MX structures, 11,294 3DEM structures, and 13,738 NMR structures.”

4. E4 Breakthrough A.I. Makes Huge Leap Toward Solving 50-Year-Old ... Smithsonian Magazine · 2 December 2020

   “With meticulous lab work, scientists have figured out the precise, 3-D shapes of about 170,000 proteins—but there are at least 200 million more to go”

The RCSB PDB 2021 Annual Report puts the archive at 185,541 entries as of January 1, 2022 (E2). That date falls in the 2021–2022 window and the figure is inside the 150,000–190,000 band. Hassabis's Nobel lecture says about 170,000 experimentally determined PDB structures were used to train AlphaFold (E1). Smithsonian reported in late 2020 that lab work had determined the 3-D shapes of about 170,000 proteins (E4). These sources are independent of the speaker and are consistent with a round figure of about 170,000 experimentally solved structures around that time; E4 dates from 2020, E1 gives no year, and E2 counts all PDB entries rather than only protein structures.