グラフ構造データに対する深層学習〜創薬・材料科学への応用とその問題点〜 (第26回ステアラボ人工知能セミナー)
6 likes4,033 views
The document discusses the work of the AIST AIRC Machine Learning Team, led by Masashi Tsubaki, in applying graph neural networks (GNNs) for molecular property prediction and related tasks. It includes details about various datasets, training parameters, and GitHub resources associated with their projects, emphasizing advancements in machine learning techniques for chemistry. The team showcases models like MolecularGNN and SchNet, highlighting performance metrics and applications in chemical research.
![[Tsubaki, Tomii, and Sese, 2018 in Bioinformatics]
[Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters]
AIST AIRC machine learning team
Masashi Tsubaki](https://image.slidesharecdn.com/tsubakiseminar-191218045744/85/26-3-320.jpg)
![[Tsubaki, Tomii, and Sese, 2018 in Bioinformatics]
[Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters]
AIST AIRC machine learning team
Masashi Tsubaki](https://image.slidesharecdn.com/tsubakiseminar-191218045744/85/26-18-320.jpg)
![AIST AIRC machine learning team
Masashi Tsubaki
🤖 ( or not)
( )
( ) (end-to-end)
( )
Graph neural network (GNN)
[Scarselli+ 09; Kearnes+ 16]](https://image.slidesharecdn.com/tsubakiseminar-191218045744/85/26-20-320.jpg)
![[Tsubaki, Tomii, and Sese, 2018 in Bioinformatics]
[Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters]
AIST AIRC machine learning team
Masashi Tsubaki](https://image.slidesharecdn.com/tsubakiseminar-191218045744/85/26-26-320.jpg)
![[Tsubaki, Tomii, and Sese, 2018 in Bioinformatics]
[Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters]
AIST AIRC machine learning team
Masashi Tsubaki](https://image.slidesharecdn.com/tsubakiseminar-191218045744/85/26-35-320.jpg)
グラフ構造データに対する深層学習〜創薬・材料科学への応用とその問題点〜 (第26回ステアラボ人工知能セミナー)
- 2. AIST AIRC machine learning team Masashi Tsubaki (3 ) ( )
- 3. [Tsubaki, Tomii, and Sese, 2018 in Bioinformatics] [Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters] AIST AIRC machine learning team Masashi Tsubaki
- 4. AIST AIRC machine learning team 1 ( ) Masashi Tsubaki 🤭 🤐 🤫 1 publish … ( ) ( )
- 5. AIST AIRC machine learning team Masashi Tsubaki https://github.com/masashitsubaki ( !)
- 6. AIST AIRC machine learning team GitHub ( SMILES ) Masashi Tsubaki https://github.com/masashitsubaki/molecularGNN_smiles GNN CC(=O)OC1=CC=CC=C1C(=O)O SMILES SMILES (※ SMILES RDKit ) (e.g., or not)
- 7. AIST AIRC machine learning team SMILES Masashi Tsubaki (SMILES 0 or 1) (SMILES ) (※GitHub )
- 8. AIST AIRC machine learning team GitHub ( ) Masashi Tsubaki Atom x y z O 0.03 0.98 0.008 H 0.06 0.02 0.002 H 0.87 1.30 0.0007 Water molecule GNN (e.g., ) https://github.com/masashitsubaki/molecularGNN_3Dstructure
- 9. AIST AIRC machine learning team Masashi Tsubaki Molecular property types Data index Each atom and its 3D coordinate in the molecule CH4 Properties in order of the above types In the following, each data is described with the same format ( README)
- 10. AIST AIRC machine learning team Masashi Tsubaki ( ) bash train.sh train.sh
- 11. AIST AIRC machine learning team bash train.sh (Google colab ) Masashi Tsubaki
- 12. AIST AIRC machine learning team (QM9 ) Masashi Tsubaki dataset: QM9_under14atoms property: U0(kcal/mol) dim: 200 layer_hidden: 6 layer_output: 6 batch_train: 32 batch_test: 32 learning_rate: 1e-3 decay of learning rate: 0.99 interval of decay: 10 iteration: 3000 1.9 kcal/mol 1.0 kcal/mol
- 13. AIST AIRC machine learning team Masashi Tsubaki data_train public data_test in house dataset 2 (e.g., ) (e.g., )
- 14. AIST AIRC machine learning team preprocess.py train.py ( ) Masashi Tsubaki ( )
- 15. AIST AIRC machine learning team Masashi Tsubaki DeepChem Schnet Chainer chemistry kGCN
- 16. AIST AIRC machine learning team Masashi Tsubaki [email protected] GitHub 🙇 …
- 17. 🙇 AIST AIRC machine learning team Masashi Tsubaki https://github.com/masashitsubaki ( !)
- 18. [Tsubaki, Tomii, and Sese, 2018 in Bioinformatics] [Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters] AIST AIRC machine learning team Masashi Tsubaki
- 19. 😅 🤔 COOH ( ) 1 0 1 (e.g., ) ( or not) AIST AIRC machine learning team ( + ) Masashi Tsubaki (e.g., )
- 20. AIST AIRC machine learning team Masashi Tsubaki 🤖 ( or not) ( ) ( ) (end-to-end) ( ) Graph neural network (GNN) [Scarselli+ 09; Kearnes+ 16]
- 21. (1) ( ) (2) (or ) ( ) (3) ( ) O C H H AIST AIRC machine learning team Masashi Tsubaki : GNN
- 22. NN ( ) ( transition propagation( ) message passing ) O C H H AIST AIRC machine learning team Masashi Tsubaki : GNN x (`+1) i = x (`) i + X j f(x (`) j ) e.g, ReLu(Wx+b) NN (4) (5)
- 23. sum or not ( ) AIST AIRC machine learning team Masashi Tsubaki O C H H
- 24. sum or not (NN ) end-to-end AIST AIRC machine learning team Masashi Tsubaki O C H H
- 25. PyTorch-like code AIST AIRC machine learning team ( ) Masashi Tsubaki C O H fif 1 else 0 H
- 26. [Tsubaki, Tomii, and Sese, 2018 in Bioinformatics] [Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters] AIST AIRC machine learning team Masashi Tsubaki
- 27. or not GNN GNN assumption ( ) Masashi Tsubaki AIST AIRC machine learning team
- 28. Masashi Tsubaki AIST AIRC machine learning team
- 29. β-lactam Masashi Tsubaki AIST AIRC machine learning team
- 30. 👍 x1 x2 xM r (e.g., r=2) Masashi Tsubaki x (`+1) i = x (`) i + X j f(x (`) j ) i-th fingerprint vector Neighboring fingerprint vector fingerprint based-GNN AIST AIRC machine learning team β-lactam Fingerprint vectors
- 31. Masashi Tsubaki ( GitHub) AIST AIRC machine learning team https://github.com/masashitsubaki/CPI_prediction fingerprint-GNN CNN
- 32. Masashi Tsubaki (human C.elegans) AIST AIRC machine learning team ( …) Dataset: Human Radius of subgraphs (fingerprints): 2 Amino acid n-gram :3 Dimensionality :10 Layer of GNN: 3 Window size of n-gram :5 Layer of CNN: 3 Learning rate: 1e-4 Decay of learning rate: 0.5 Interval of decay: 10 Dataset: C.elegans Radius of subgraphs (fingerprints): 2 Amino acid n-gram :3 Dimensionality :10 Layer of GNN: 3 Window size of n-gram :5 Layer of CNN: 3 Learning rate: 1e-4 Decay of learning rate: 0.5 Interval of decay: 10 Liu et al., 2015 - (highly credible negative samples)
- 33. Masashi Tsubaki AIST AIRC machine learning team ... some reports even above 0.99 AUC on standard benchmarks…
- 34. Masashi Tsubaki AIST AIRC machine learning team ( ) …
- 35. [Tsubaki, Tomii, and Sese, 2018 in Bioinformatics] [Tsubaki and Mizoguchi, 2018 in Journal of physical chemistry letters] AIST AIRC machine learning team Masashi Tsubaki
- 36. AIST AIRC machine learning team Masashi Tsubaki Atom x y z O 0.03 0.98 0.008 H 0.06 0.02 0.002 H 0.87 1.30 0.0007 3 ( ) -9.24 eV Water molecule ( …) ( …) ( …)
- 37. AIST AIRC machine learning team Nature comm PRL ( …) Masashi Tsubaki ( )
- 38. AIST AIRC machine learning team Masashi Tsubaki ( …) ( ) O C H H ※
- 39. AIST AIRC machine learning team Masashi Tsubaki Schnet( ) i x (`+1) i = X j e j d2 ij f(x (`) j ) j ( …) ( )
- 40. AIST AIRC machine learning team Masashi Tsubaki Gaussian ( ) = O H H f ⌘ exp ⇣ 2 ※
- 41. AIST AIRC machine learning team Masashi Tsubaki (※QM9 29 130k ) QM9 14 (15k ) ( )
- 42. AIST AIRC machine learning team Masashi Tsubaki QM9 14 (10k ) 1.0 Chemical accuracy 1.90 kcal/mol
- 43. AIST AIRC machine learning team Masashi Tsubaki 1 ( ) 1.24 kcal/mol GNN ( …) 1.90 kcal/mol 1.0
- 44. AIST AIRC machine learning team Masashi Tsubaki New! GNN ( )
- 45. AIST AIRC machine learning team ( ) Masashi Tsubaki 14 15 New!
- 46. AIST AIRC machine learning team Masashi Tsubaki Atom x y z O 0.03 0.98 0.008 H 0.06 0.02 0.002 H 0.87 1.30 0.0007 ( ) = 1 -9.24 eV Water molecule = etc… … ( ) 🤭
- 47. AIST AIRC machine learning team Masashi Tsubaki SchNOrb extends the deep tensor neural network SchNet to represent electronic wavefunctions, ...model uses about 93 million parameters to predict a large Hamiltonian…
- 48. AIST AIRC machine learning team Masashi Tsubaki https://github.com/masashitsubaki [email protected]