Development of Machine Learning Models withFragment Molecular Orbital Calculation Data

Abstract

フラグメント分子軌道(FMO)法はタンパク質全体を量子化学計算可能な稀有な手法である．そして，FMO法によって得られるデータもまた，現状ではタンパク質系の量子化学計算データとして唯一無二のものとなっている．汎用ソフトウェアでは生成が困難なタンパク質の量子化学計算データとそれを用いた様々な機械学習モデルの開発は，近年活性化が著しいAI創薬に大きなインパクトを与えることが期待される．本稿では，筆者らのグループで進めているFMOデータを用いた機械学習モデル(原子電荷予測モデル，相互作用予測モデル，機械学習力場)の開発状況を概説する．

Translated Abstract

Abstract: Fragment Molecular Orbital (FMO) is a unique method that allows quantum mechanical (QM) calculations of entire proteins. The data obtained by the FMO method are also currently the only QM calculation data for protein systems. The development of various machine learning models using the QM calculation data of proteins, which are difficult to generate with general-purpose software, is expected to have a significant impact on AI drug discovery, which has been remarkably active in recent years. This paper outlines the status of the development of machine learning models using FMO data, which is ongoing in the author's group.

Figures

Figure 1

 Comparison of FMO calculated and NN predicted values for RESP charges for the structure with the highest coefficient of determination (R²) in the Test data.

Figure 2

 Comparison of FMO calculated and NN predicted values of IFIE-Sum for 20 poses of each of 15 Test ligands of CDK2 (total 300 data).

Figure 3

 Benchmark of machine learning forcefield training scripts for water-only system.

Tables

Table 1 Details of the dataset used to build the atomic charge prediction model. Note that the numbers in the table do not include water molecules.

	polyQ10	TrpCage	BRD2
PDBID	2OTU	1L2Y	5IBN
Residues (Atoms)	10(173)	20(304)	111(1,811)
MD snapshots	10,000	10,000	1,000
FMO data	10,000	9,805	902
Train	7,997	7,840	718
Validation	2,000	1,961	180
Test	3+1,000	4+989	4+180

Table 2 The coefficient of determination (R²) and the net charge for test data at the cluster center for each target are presented. The net charge represents the sum of the atomic charges of all atoms. The correct values for polyQ10, TrpCage, and BRD2 are 0, 1, and 1, respectively.

	polyQ10		TrpCage		BRD2
Test	R2	Net charge	R2	Net charge	R2	Net charge
1	0.982	-0.31	0.928	1.50	0.896	1.35
2	0.960	-0.23	0.926	0.97	0.888	3.63
3	0.955	0.46	0.917	0.51	0.886	-0.48
4			0.900	1.38	0.873	4.60

参考文献

• [1]   W. D. Cornell, P. Cieplak, C. I. Bayly, I. R. Gould, K. M. Merz, D. M. Ferguson, et al., J. Am. Chem. Soc., 117, 5179 (1995). doi:10.1021/ja00124a002
• [2]   W. L. Jorgensen, J. Tirado-Rives, J. Am. Chem. Soc., 110, 1657 (1988). doi:10.1021/ja00214a001 PMID:27557051
• [3]   A. D. MacKerell, Jr., N. Banavali, N. Foloppe, Biopolymers, 56, 257 (2000-2001). PMID:11754339
• [4]   C. Oostenbrink, A. Villa, A. E. Mark, W. F. Van Gunsteren, J. Comput. Chem., 25, 1656 (2004). doi:10.1002/jcc.20090 PMID:15264259
• [5]   J. Behler, M. Parrinello, Phys. Rev. Lett., 98, 146401 (2007). doi:10.1103/PhysRevLett.98.146401 PMID:17501293
• [6]   S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, et al., Nat. Commun., 13, 2453 (2022). doi:10.1038/s41467-022-29939-5 PMID:35508450
• [7]   S. Takamoto, S. Izumi, J. Li, Comput. Mater. Sci., 207, 111280 (2022). doi:10.1016/j.commatsci.2022.111280
• [8]   A. P. Bartók, M. C. Payne, R. Kondor, G. Csányi, Phys. Rev. Lett., 104, 136403 (2010). doi:10.1103/PhysRevLett.104.136403 PMID:20481899
• [9]   G. Kresse, J. Furthmüller, Phys. Rev. B Condens. Matter, 54, 11169 (1996). doi:10.1103/PhysRevB.54.11169 PMID:9984901
• [10]   G. Kresse, J. Furthmüller, Comput. Mater. Sci., 6, 15 (1996). doi:10.1016/0927-0256(96)00008-0
• [11]   G. Kresse, J. Hafner, Phys. Rev. B Condens. Matter, 47, 558 (1993). doi:10.1103/PhysRevB.47.558 PMID:10004490
• [12]   P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, M. Calandra, et al., J. Phys. Condens. Matter, 29, 465901 (2017). doi:10.1088/1361-648X/aa8f79 PMID:29064822
• [13]   P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, et al., J. Phys. Condens. Matter, 21, 395502 (2009). doi:10.1088/0953-8984/21/39/395502 PMID:21832390
• [14]Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Petersson, G. A.; Nakatsuji, H.; et al. Gaussian 16 Rev. C.01. 2016.
• [15]   NequIP; https://github.com/mir-group/nequip.
• [16]   n2p2; https://compphysvienna.github.io/n2p2/index.html.
• [17]K. Kato, T. Masuda, C. Watanabe, N. Miyagawa, H. Mizouchi, S. Nagase, J. Chem Inf. Model. 2020, 60 (7), 3361-3368, Article. doi:10.1021/acs.jcim.0c00273
• [18]   O. T. Unke, M. Stöhr, S. Ganscha, T. Unterthiner, H. Maennel, S. Kashubin, et al., Sci. Adv., 10, eadn4397 (2024). doi:10.1126/sciadv.adn4397 PMID:38579003
• [19]   K. Kitaura, E. Ikeo, T. Asada, T. Nakano, M. Uebayasi, Chem. Phys. Lett., 313, 701 (1999). doi:10.1016/S0009-2614(99)00874-X
• [20]   J. Behler, J. Chem. Phys., 134, 074106 (2011). doi:10.1063/1.3553717 PMID:21341827
• [21]   M. Gastegger, L. Schwiedrzik, M. Bittermann, F. Berzsenyi, P. Marquetand, J. Chem. Phys., 148, 241709 (2018). doi:10.1063/1.5019667 PMID:29960372
• [22]   N. Artrith, A. Urban, G. Ceder, Phys. Rev. B, 96, 014112 (2017). doi:10.1103/PhysRevB.96.014112
• [23]   FMO Drug Design Consortium. https://fmodd.jp/
• [24]   Y. Mochizuki, T. Nakano, K. Sakakura, Y. Okiyama, H. Watanabe, K. Kato, et al., The ABINIT-MP Program. In Recent Advances of the Fragment Molecular Orbital Method, SpringerLink, 2021; pp 53-67.
• [25]S. Tanaka, Y. Mochizuki, Y. Komeiji, Y. Okiyama, K. Fukuzawa, Phys. Chem. Chem. Phys., 2014, 16 (22), 10310-10344. doi:10.1039/C4CP00316K
• [26]   R. Kita, C. Watanabe, M. Ohta, N. Tanimura, K. Okuwaki, M. Ikeguchi, et al., Proceedings of the Annual Conference of JSAI 2024, JSAI2024, 4Q3IS2d05-04Q03IS02d05. doi:10.11517/pjsai.JSAI2024.0_4Q3IS2d05
• [27]   Life Intelligence Consortium. https://linc-ai.jp/
• [28]   T. Morawietz, A. Singraber, C. Dellago, J. Behler, Proc. Natl. Acad. Sci. USA, 113, 8368 (2016). doi:10.1073/pnas.1602375113 PMID:27402761
• [29]   S. Tokutomi, K. Shimamura, K. Fukuzawa, S. Tanaka, Chem. Phys. Lett., 757, 137883 (2020). doi:10.1016/j.cplett.2020.137883
• [30]   H. Doi, S. Matsuoka, K. Okuwaki, R. Hatada, S. Minami, R. Suhara, et al., Jpn. J. Appl. Phys., 62, 070901 (2023). doi:10.35848/1347-4065/ace575