DeepMind publishes AlphaFold's method and code, and had filed its patent applications before doing so

2021-07-15Industry

John Jumper et al., "Highly accurate protein structure prediction with AlphaFold," Nature 596, 583–589 (published 15 July 2021), with source code, trained weights and inference script released at github.com/deepmind/alphafold

DeepMind published the AlphaFold methods paper in Nature, received 11 May 2021 and accepted 12 July 2021, and released the model on the same day. The paper reports that in the fourteenth Critical Assessment of protein Structure Prediction, AlphaFold structures had a median backbone accuracy of 0.96 Å r.m.s.d.95 against 2.8 Å for the next best method, and describes the system as the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known. The code availability statement says that source code for the AlphaFold model, trained weights and inference script are available under an open-source license at github.com/deepmind/alphafold, and the repository licenses the code under the Apache License 2.0. The competing interests statement discloses that eleven of the authors, including Jumper, Evans, Pritzel, Green, Figurnov, Ronneberger, Bates, Bridgland, Kohl, Reiman and Senior, have filed non-provisional patent applications 16/701,070 and PCT/EP2020/084238, together with several provisional applications, with DeepMind Technologies Limited, pending and relating to machine learning for predicting protein structures. The United States application was filed on December 2, 2019, before the model's CASP14 results in late 2020 and before this publication.

What it changesThe sequence matters more than the science for anyone who holds or seeks patents on models. DeepMind filed in December 2019, competed in CASP14 in 2020, and published the method, the code and the weights in July 2021, so the public disclosure came after the priority date and the applications are unaffected by it. A company that reverses that order and releases weights or code first has, from the release date, one year to file in the United States under the grace period in Section 102(b)(1)(A) of Title 35 of the United States Code and no path at all in Europe or China for anything the release makes obvious. The release also becomes prior art against every protein-structure application filed after it. Before any model release, the program should have a filing on the architecture and training method, a decision on which engineering details stay out of the repository, and a check of the open-source license for its patent grant, because the Apache License 2.0 carries one.

More on Prior art, public disclosure and obviousness · Trade secrets and open release

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.