Human-in-the-loop assisted de novo molecular design

Journal of Cheminformatics

Research

Haoping Xiao¹  Markus Heinonen¹  Iiris Sundin¹  Samuel Kaski²  Alexey Voronov³  Ola Engkvist⁴  Esben Jannik Bjerrum⁵  Atanas Patronov⁵  Kostas Papadopoulos⁵ 

[1] Department of Computer Science, Aalto University, Espoo, Finland;Department of Computer Science, Aalto University, Espoo, Finland;Department of Computer Science, University of Manchester, Manchester, UK;Molecular AI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden;Molecular AI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden;Department of Computer Science and Engineering, Chalmers University of Technology, Gothenburg, Sweden;Molecular AI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden;Odyssey Therapeutics, Cambridge, MA, USA;

关键词: Interactive algorithms;    De novo molecular design;    Human-in-the-loop;    AI-assisted design;    Goal-oriented molecule generation;    Expert knowledge elicitation;    Reward elicitation;   

DOI  :  10.1186/s13321-022-00667-8

 received in 2022-06-30, accepted in 2022-12-03,  发布年份 2022

来源: Springer

【 摘 要 】

A de novo molecular design workflow can be used together with technologies such as reinforcement learning to navigate the chemical space. A bottleneck in the workflow that remains to be solved is how to integrate human feedback in the exploration of the chemical space to optimize molecules. A human drug designer still needs to design the goal, expressed as a scoring function for the molecules that captures the designer’s implicit knowledge about the optimization task. Little support for this task exists and, consequently, a chemist usually resorts to iteratively building the objective function of multi-parameter optimization (MPO) in de novo design. We propose a principled approach to use human-in-the-loop machine learning to help the chemist to adapt the MPO scoring function to better match their goal. An advantage is that the method can learn the scoring function directly from the user’s feedback while they browse the output of the molecule generator, instead of the current manual tuning of the scoring function with trial and error. The proposed method uses a probabilistic model that captures the user’s idea and uncertainty about the scoring function, and it uses active learning to interact with the user. We present two case studies for this: In the first use-case, the parameters of an MPO are learned, and in the second use-case a non-parametric component of the scoring function to capture human domain knowledge is developed. The results show the effectiveness of the methods in two simulated example cases with an oracle, achieving significant improvement in less than 200 feedback queries, for the goals of a high QED score and identifying potent molecules for the DRD2 receptor, respectively. We further demonstrate the performance gains with a medicinal chemist interacting with the system.Graphical Abstract

【 授权许可】

CC BY   
© The Author(s) 2022

【 图 表 】

41408_2022_764_Article_IEq25.gif

41408_2022_764_Article_IEq24.gif

41408_2022_764_Article_IEq23.gif

41408_2022_764_Article_IEq22.gif

41408_2022_764_Article_IEq21.gif

41408_2022_764_Article_IEq20.gif

41408_2022_764_Article_IEq19.gif

Fig. 1

41408_2022_764_Article_IEq18.gif

41408_2022_764_Article_IEq17.gif

41408_2022_764_Article_IEq16.gif

41408_2022_764_Article_IEq15.gif

41408_2022_764_Article_IEq14.gif

41408_2022_764_Article_IEq13.gif

41408_2022_764_Article_IEq12.gif

41408_2022_764_Article_IEq11.gif

41408_2022_764_Article_IEq10.gif

41408_2022_764_Article_IEq9.gif

41408_2022_764_Article_IEq8.gif

41408_2022_764_Article_IEq7.gif

41408_2022_764_Article_IEq6.gif

Fig. 2

41408_2022_764_Article_IEq3.gif

Fig. 1

41408_2022_764_Article_IEq1.gif

Fig. 2

Fig. 5

Fig. 1

Fig. 1

40249_2022_1045_Article_IEq27.gif

Fig. 5

40249_2022_1045_Article_IEq25.gif

40249_2022_1045_Article_IEq24.gif

40249_2022_1045_Article_IEq23.gif

40249_2022_1045_Article_IEq22.gif

40249_2022_1045_Article_IEq21.gif

40249_2022_1045_Article_IEq20.gif

40249_2022_1045_Article_IEq17.gif

40249_2022_1045_Article_IEq16.gif

Fig. 1

40249_2022_1045_Article_IEq15.gif

40249_2022_1045_Article_IEq14.gif

40249_2022_1045_Article_IEq13.gif

40249_2022_1045_Article_IEq12.gif

40249_2022_1045_Article_IEq11.gif

40249_2022_1045_Article_IEq10.gif

40249_2022_1045_Article_IEq9.gif

40249_2022_1045_Article_IEq8.gif

40249_2022_1045_Article_IEq7.gif

40249_2022_1045_Article_IEq6.gif

Fig. 1

40249_2022_1045_Article_IEq1.gif

Fig. 4

Fig. 1

Fig. 1

Fig. 4

12864_2022_9026_Article_IEq26.gif

Fig. 3

Fig. 3

Fig. 3

Fig. 8

Fig. 2

Fig. 2

Fig. 2

Fig. 1

Fig. 1

Fig. 2

Fig. 6

Fig. 8

Fig. 1

Fig. 2

Fig. 3

Fig. 6

Fig. 2

Fig. 3

Fig. 3

Fig. 3

Fig. 2

Fig. 2

Fig. 3

13731_2022_257_Article_IEq12.gif

Fig. 1

13731_2022_257_Article_IEq9.gif

13731_2022_257_Article_IEq7.gif

Fig. 2

Fig. 1

13731_2022_257_Article_IEq4.gif

Fig. 3

13731_2022_257_Article_IEq2.gif

Fig. 3

Fig. 1

Fig. 6

Fig. 5

Fig. 2

Fig. 2

Fig. 5

Fig. 1

Fig. 1

Fig. 1

Fig. 1

40708_2022_178_Article_IEq64.gif

Fig. 2

Fig. 1

Fig. 1

Fig.5

Fig. 5

Fig. 4

Fig. 2

Fig. 3

Fig. 5

40708_2022_178_Article_IEq53.gif

Fig. 2

Fig. 1

Fig. 6

Fig. 4

Fig. 1

Fig. 1

Fig. 4

Fig. 1

Fig. 7

40708_2022_178_Article_IEq42.gif

Fig. 3

Fig. 6

40708_2022_178_Article_IEq39.gif

Fig. 6

Fig. 2

40708_2022_178_Article_IEq35.gif

40517_2022_243_Article_IEq20.gif

40517_2022_243_Article_IEq19.gif

Fig. 1

Fig. 4

Fig. 2

Fig. 3

40708_2022_178_Article_IEq25.gif

Fig. 2

Fig. 3

40517_2022_243_Article_IEq17.gif

40517_2022_243_Article_IEq16.gif

Fig. 2

Fig. 1

40708_2022_178_Article_IEq17.gif

Fig. 3

Fig. 5

Fig. 1

40517_2022_243_Article_IEq14.gif

Fig. 2

40517_2022_243_Article_IEq12.gif

40517_2022_243_Article_IEq11.gif

Fig. 1

12864_2022_9026_Article_IEq3.gif

Fig. 2

Fig. 4

Fig. 1

Fig. 2

Fig. 3

40517_2022_243_Article_IEq10.gif

Fig. 4

Fig. 9

Fig. 3

Fig. 3

Fig. 4

Fig. 4

40517_2022_243_Article_IEq3.gif

Fig. 2

Fig. 1