US 12,437,843 B2
Predicting protein structures using geometry neural networks that estimate similarity between predicted protein structures and actual protein structures
Andrew W. Senior, London (GB); James Kirkpatrick, London (GB); Laurent Sifre, Paris (FR); Richard Andrew Evans, London (GB); Hugo Penedones, Zurich (CH); Chongli Qin, London (GB); Ruoxi Sun, Mountain View, CA (US); Karen Simonyan, London (GB); and John Jumper, London (GB)
Assigned to GDM Holding LLC, Mountain View, CA (US)
Appl. No. 17/266,724
Filed by DeepMind Technologies Limited, London (GB)
PCT Filed Sep. 16, 2019, PCT No. PCT/EP2019/074676
§ 371(c)(1), (2) Date Feb. 8, 2021,
PCT Pub. No. WO2020/058177, PCT Pub. Date Mar. 26, 2020.
Claims priority of provisional application 62/770,490, filed on Nov. 21, 2018.
Claims priority of provisional application 62/734,773, filed on Sep. 21, 2018.
Claims priority of provisional application 62/734,757, filed on Sep. 21, 2018.
Prior Publication US 2021/0304847 A1, Sep. 30, 2021
Int. Cl. G16B 40/20 (2019.01); G06F 18/2413 (2023.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06N 3/08 (2023.01); G06N 20/00 (2019.01); G16B 15/20 (2019.01); G16H 10/40 (2018.01)
CPC G16B 40/20 (2019.02) [G06F 18/24147 (2023.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06N 3/08 (2013.01); G16B 15/20 (2019.02); G16H 10/40 (2018.01); G06N 20/00 (2019.01)] 18 Claims
 
1. A method comprising:
obtaining a ligand, wherein the ligand is a drug for treating a disease, wherein the ligand is an agonist or antagonist of a protein, wherein the protein is a receptor or enzyme, wherein obtaining the ligand comprises performing, by one or more computers, operations including:
determining a predicted structure of the protein, comprising:
performing a plurality of update iterations, comprising, at each update iteration:
maintaining data including: (i) a current predicted structure of the protein defined by current values of a plurality of structure parameters, and (ii) a quality score characterizing a quality of the current predicted structure based on a current geometry score that is an estimate of a similarity measure between the current predicted structure and an actual structure of the protein;
determining an alternative predicted structure of the protein based on the current predicted structure, wherein the alternative predicted structure is defined by alternative values of the structure parameters;
processing, using a geometry neural network and in accordance with current values of geometry neural network weights, a network input comprising: (i) a representation of a sequence of amino acid residues in the protein, and (ii) the alternative values of the structure parameters, to generate, as an output of the geometry neural network, a numerical value defining an alternative geometry score that is an estimate of a similarity measure between the alternative predicted structure and the actual structure of the protein;
determining a quality score characterizing a quality of the alternative predicted structure based on the alternative geometry score; and
determining whether to update the current predicted structure to the alternative predicted structure using the quality score characterizing the quality of the current predicted structure and the quality score characterizing the quality of the alternative predicted structure; and
determining the predicted structure of the protein to be defined by values of the plurality of structure parameters that are associated with at least a threshold quality score;
evaluating an interaction of one or more candidate ligands with the predicted structure of the protein; and
selecting one or more of the candidate ligands as the ligand dependent on a result of the evaluating, comprising selecting the one or more candidate ligands that are predicted to bind to the protein with sufficient affinity for a biological effect; and
synthesizing the one or more candidate ligands selected as the ligand.