用于高精度催化性能预测的精细结构敏感型深度学习框架

doi:10.1016/S1872-2067(23)64467-5

催化学报 ›› 2023, Vol. 50: 284-296.DOI: 10.1016/S1872-2067(23)64467-5

用于高精度催化性能预测的精细结构敏感型深度学习框架

陈宇卓^a, 王浩^a, 陆冰^a, 易倪^b, 曹亮^b, 王勇^a, 毛善俊^a^,^*()

^a浙江大学化学系, 催化研究所先进材料与催化课题组, 浙江杭州 310028
^b浙江大学化学系, 催化研究所, 浙江杭州 310028

收稿日期:2023-03-20 接受日期:2023-06-01 出版日期:2023-07-18 发布日期:2023-07-25
通讯作者: *电子信箱: maoshanjun@zju.edu.cn (毛善俊).
基金资助:
国家重点研发计划(2021YFB3801600);浙江省“尖兵” “领雁”研发攻关计划项目(2022C01218);浙江省“尖兵” “领雁”研发攻关计划项目(2022C01151);浙江省“尖兵” “领雁”研发攻关计划项目(2023C01108)

Fine-structure sensitive deep learning framework for predicting catalytic properties with high precision

Yuzhuo Chen^a, Hao Wang^a, Bing Lu^a, Ni Yi^b, Liang Cao^b, Yong Wang^a, Shanjun Mao^a^,^*()

^aAdvanced Materials and Catalysis Group, Center of Chemistry for Frontier Technologies, State Key Laboratory of Clean Energy Utilization, Institute of Catalysis, Department of Chemistry, Zhejiang University, Hangzhou 310028, Zhejiang, China
^bInstitute of Catalysis, Department of Chemistry, Zhejiang University, Hangzhou 310028, Zhejiang, China

Received:2023-03-20 Accepted:2023-06-01 Online:2023-07-18 Published:2023-07-25
Contact: *E-mail: maoshanjun@zju.edu.cn (S. Mao).
Supported by:
National Key R&D Program of China(2021YFB3801600);“Pioneer” and “Leading Goose” R&D Program of Zhejiang Province(2022C01218);“Pioneer” and “Leading Goose” R&D Program of Zhejiang Province(2022C01151);“Pioneer” and “Leading Goose” R&D Program of Zhejiang Province(2023C01108)

摘要/Abstract

摘要：

催化剂表面的精细结构对结构敏感型反应有很大影响, 高通量(HT)筛选和机器学习(ML)可以有效地探索这些影响因素. 为了将ML与化学相结合, 必须首先将化学结构转换为可用作ML模型输入的特征编码, 目前常用的两种转换方法为描述符和图. 然而, 描述符的构建往往忽略原子连接, 这使得ML模型难以捕获与催化性能最相关的几何信息. 基于图的ML模型在更新节点的过程中会不可避免地丢失吸附位点的几何排列信息, 同时消息传递神经网络复杂, 导致其对电子或几何结构不敏感、缺乏可解释性. 因此, 目前仍然缺乏可以同时兼顾多相催化中电子和几何精细结构的可解释ML框架.

相比之下, 将化学结构转换为网格数据可以完全保留精细的几何信息. 鉴于此, 通过将催化剂表面结构和吸附位点信息分别转换为二维网格和一维描述符, 本文创建了一个名为“整体+局部”卷积神经网络(GLCNN)的数据增强(DA)卷积神经网络(CNN)ML框架, 其结合“整体+局部”特征, 无需复杂的编码即可捕获原始精细结构, DA的加入可以扩充数据集并减缓过拟合. GLCNN可以很好地预测和区分碳基过渡金属单原子催化剂上OH的吸附能, 平均绝对误差小于0.1 eV, 这是在大型数据集上训练的ML模型所能达到的较好结果. 将GLCNN与基于描述符或图的模型对比, 结果表明, 对比模型无法完全准确预测包含IB和IIB过渡金属或者顺式/反式构型催化剂的OH吸附能. 而GLCNN模型的预测效果明显好于对比模型, 表明网格和描述符的组合可以更好地体现催化活性中心的电子和精细几何结构信息. 另外, 对DA处理后的样本计算平均标准误差后发现, 通过DA获得的不同晶胞几乎不影响预测结果, 说明DA对晶胞的平移并不改变晶胞的性质, 表明GLCNN可以学习到周期性表面的边界条件信息. 与传统的CNN和基于描述符的单边特征提取不同, 本文中对精细结构敏感的ML框架可以通过不包含人类偏见的可解释性分析, 从几何和化学/电子特征中提取影响催化性能的关键因素, 如对称和配位元素. 一维描述符的特征重要性分析表明, 吸附位点的电子结构和对称性特征至关重要, 且金属对于催化性能的影响强于其配位环境. 将CNN卷积部分的中间输出可视化后发现, 碳基载体上远离金属的区域中很大一部分对催化性能几乎没有直接影响, 且卷积层会优先对金属原子反复关注, 再次强调金属的重要性高于其配位环境, 表明卷积核可以自动提取符合催化常识化学结构的几何信息. 对全连接层(FC)进行降维可视化分析后发现, 随着层数的增加, FC基于基本催化知识逐渐寻找特征提取的方向, 提取出更抽象的有利于吸附能预测的高维特征, 这与卷积部分类似. 综上, GLCNN框架为具有广阔物理和化学空间的多相催化剂的高精度HT筛选提供了可行方案.

关键词: 非均相催化, 机器学习, 精细结构敏感, “整体+局部”特征, 可解释性, 数据增强, “整体+局部”卷积神经网络

Abstract:

The fine structure of a surface considerably affects its catalytic performance in structurally sensitive reactions. High-throughput (HT) screening and machine learning (ML) are considered efficient for exploring the hidden rules of impacts. However, no protocol for constructing an interpretable ML framework sensitive to fine structures has been reported thus far. Herein, we developed a data augmented convolutional neural network (CNN)-based ML framework called "global + local" convolutional neural network (GLCNN), which combines "global + local" features. This framework captures original fine structures without the use of complicated encoding methods by transforming the catalytic surfaces and adsorption sites into two-dimensional grids and one-dimensional descriptors, respectively. The GLCNN framework accurately predicted and distinguished the adsorption energies of OH on a set of analogous carbon-based transition-metal single-atom catalysts with a mean absolute error of less than 0.1 eV. Moreover, this model yields the best results among popular models trained on large datasets so far. Unlike conventional CNN and descriptor-based models with one-sided feature extraction, this fine-structure-sensitive ML framework can extract key factors that affect the catalytic performance from both geometric and chemical/electronic features, such as symmetry and coordination elements, through unbiased interpretable analysis. This framework provides a feasible solution for the high-precision HT screening of heterogeneous catalysts with a broad physical and chemical space.

Key words: Heterogeneous catalyst, Machine learning, Fine-structure sensitive, "Global + local" feature, Interpretability, Data augmentation, "Global + local" convolutional neural network

陈宇卓, 王浩, 陆冰, 易倪, 曹亮, 王勇, 毛善俊. 用于高精度催化性能预测的精细结构敏感型深度学习框架[J]. 催化学报, 2023, 50: 284-296.

Yuzhuo Chen, Hao Wang, Bing Lu, Ni Yi, Liang Cao, Yong Wang, Shanjun Mao. Fine-structure sensitive deep learning framework for predicting catalytic properties with high precision[J]. Chinese Journal of Catalysis, 2023, 50: 284-296.

×
扫码分享

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.cjcatal.com/CN/10.1016/S1872-2067(23)64467-5

https://www.cjcatal.com/CN/Y2023/V50/I7/284

图/表 11

Fig. 1. Overview of the descriptor (a), grid (b), graph (c), and the "global + local" (d) representations of the chemical structure used for ML, where CNN and MPNN denote convolution neural network and message passing neural network [38], respectively. The white, red, blue and gray squares illustrate the feature vectors corresponding to the different representations.

Table 1 MAE* and MAE for predicting OH adsorption energies (unit: eV).

DA times ^a	Total ^*	Grid ^*	Descriptor ^*
0	0.134^b (NA)^c	0.159 (NA)	0.361 (NA)
5	0.126 (0.120)	0.139 (0.134)	NA (NA)
10	0.113 (0.106)	0.135 (0.130)	NA (NA)
15	0.105 (0.100)	0.131 (0.126)	NA (NA)
18	0.104 (0.100)	0.130 (0.126)	NA (NA)
20	0.103 (0.099)	0.129 (0.125)	NA (NA)

Table 2 Benchmark comparison of models for predicting OH chemisorption.

Algorithm	#Sample	Range (eV)	MAE (eV)	RMSE (eV)	Ref.
ANN ^b	635	1.8	NA	0.240	[68]
ANN	748	4.8	0.152	0.222	[67]
GPR ^c	1235	4.6	0.170	0.240	[69]
Bayesian	512	2.2	0.160	0.209	[70]
Bayesian	748	4.8	0.270	0.435	[67]
DOSnet	1103	5.4	0.156	0.221	[71]
CGCNN	748	4.8	0.114	0.189	[67]
TinNet	748	4.8	0.118	0.188	[67]
LR ^d	112	5.6	0.210	NA	[12]^a
GLCNN	1564^e	5.6	0.114 (0.110)^*	0.188 (0.182)^*	This work ^a
GLCNN	3128	5.6	0.103 (0.099)^*	0.173 (0.167)^*	This work ^a

Fig. 2. Symmetry elements contained in SV (a), DV (b) and HV (c) defects with different N contents, where Cn, i, σ and E represent n-fold rotation axis, symmetry center, vertical mirror, and identity element, respectively. Numbers in parentheses distinguish structures containing different symmetry elements. TM atom protrudes from the plane in SV, while in DV and HV, it is in the same plane as other atoms owing to the large defect space in DV and HV.

Fig. 3. Workflow of the double-input single-output GLCNN framework based on the "global + local" strategy, where the gray and blue spheres represent C and Pd atoms, respectively. First, the original chemical structure is expanded and gridded, and multiple channels are added. Subsequently, all channels are cropped, flipped, and resized to generate the input of the CNN module. In contrast, the 1D descriptor related to the adsorption site is fed as the second input and concatenated with the output from the CNN module to form a new input, which is finally transmitted to the fully connected (FC) module to obtain the predicted adsorption energy.

Fig. 4. Structure of original sample (a) and a collection of (b) cropped and flipped data (c) derived from it, taking the channel of atomic number as example. Electronegativity channels of catalysts with (d) cis and (e) trans configurations. The original grid with a size of 128×128 is reduced to 32×32 after the processing of a 64×64 sliding window and Gaussian reconstruction. Gray and blue spheres represent C and Pd atoms, respectively. All augmented data represents the same chemical structure. The dark blue background and fluorescent green/blue highlights represent the vacuum area and atoms respectively.

Fig. 5. Architecture of the CNN model. Grids and descriptors are the two inputs of GLCNN, where 6×(32×32) represents grids with the size of 32×32 containing 6 channels, and 15×1 denotes 15 features. Conv2d represents a 2D convolution layer, where 6×(5×5) represents 6 kernels with the size of 5×5. MaxPool2d is the maximum pooling layer with the size of the pooling window set at (2×2). FC is the fully connected layer, where 101×2000 represents the shape of the input and output of the FC layer. The blue and red boxes belong to “global” and “local” sections, respectively. Layers in the green box integrate the outputs of the "global" and "local" sections and produce the final prediction.

Fig. 6. (a) Learning curve of the GLCNN in predicting OH adsorption energies, in which MAE*s are estimated by the five-fold cross-validation approach. Comparison between DFT-calculated and model-predicted values of OH adsorption energies using GLCNN (b), linear regression13 (c) and CGCNN35 (d), where a histogram of data distribution is displayed on the top of (b). Insets show the comparison between DFT-calculated and model-predicted values of data with cis/trans (red/green) configurations in the test set. Points of the same shape denote paired cis/trans configurations with the same cell expansion coefficient, defect type, and N content.

Fig. 7. (a) PFI analysis of the GLCNN model. The black vertical dashed line represents the MAE* of the model without permutation (0.099 eV). Distribution of the difference of OH adsorption energies for DV (b) and HV (c) defects with cis/trans configurations. Energy subtraction occurs between cis and trans configurations with the same cell expansion coefficient, defect type, and N content (i.e., Fig. S1(a) and Fig. S1(b), or Fig. S1(c) and Fig. S1(d)). A difference greater than 0.6 eV is rare and not shown.

Fig. 8. Intermediate output of the first convolution (a) and activation (b) layer and the second convolution (c) and activation layer (d). The red circle enclosed the obvious feature extraction. Thermal diagrams obtained from Grad-CAM analysis of the first (e) and second (f) convolution layer.

Fig. 9. t-SNE analysis using the adsorption energy of OH (a,d,g), group number of TM (b,e,h), defect type (c,f,i) as fitting or classification criteria of the first (a?c), second (d?f), and third (g?i) FC layers, respectively. The numbers 3-12 in (b) represent the group of the TMs in the periodic table from left to right, which is consistent with the element attribute setting in the pymatgen python package (Fig. S3).

参考文献

[1]	T. Bligaard, J. K. Norskov, S. Dahl, J. Matthiesen, C. H. Christensen, J. Sehested, J. Catal. 2004, 224, 206-217.
[2]	A. A. Balandin, Adv. Catal. 1969, 19, 1-210.
[3]	C. J. H. Jacobsen, S. Dahl, B. S. Clausen, S. Bahn, A. Logadottir, J. K. Nørskov, J. Am. Chem. Soc. 2001, 123, 8404-8405.
[4]	E. Skulason, T. Bligaard, S. Gudmundsdottir, F. Studt, J. Rossmeisl, F. Abild-Pedersen, T. Vegge, H. Jonsson, J. K. Nørskov, Phys. Chem. Chem. Phys. 2012, 14, 1235-1245.
[5]	A. A. Latimer, A. R. Kulkarni, H. Aljama, J. H. Montoya, J. S. Yoo, C. Tsai, F. Abild-Pedersen, F. Studt, J. K. Nørskov, Nat. Mater. 2017, 16, 225-229.
[6]	J. Greeley, I. E. L. Stephens, A. S. Bondarenko, T. P. Johansson, H. A. Hansen, T. F. Jaramillo, J. Rossmeisl, I. Chorkendorff, J. K. Nørskov, Nat. Chem. 2009, 1, 552-556.
[7]	J. Greeley, T. F. Jaramillo, J. Bonde, I. Chorkendorff, J. K. Nørskov, Nat. Mater. 2006, 5, 909-913.
[8]	J. Rossmeisl, Z. W. Qu, H. Zhu, G. J. Kroes, J. K. Nørskov, J Electroanal Chem, 2007, 607, 83-89.
[9]	J. K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J. R. Kitchin, T. Bligaard, H. Jonsson, J. Phys. Chem. B, 2004, 108, 17886-17892.
[10]	J. K. Nørskov, T. Bligaard, A. Logadottir, J. R. Kitchin, J. G. Chen, S. Pandelov, J. K. Nørskov, J. Electrochem. Soc. 2005, 152, J23-J26.
[11]	A. Kulkarni, S. Siahrostami, A. Patel, J. K. Nørskov, Chem. Rev. 2018, 118, 2302-2312.
[12]	H. Xu, D. Cheng, D. Cao, X. C. Zeng, Nat. Catal. 2018, 1, 339-348.
[13]	I. C. Man, H. Y. Su, F. Calle-Vallejo, H. A. Hansen, J. I. Martinez, N. G. Inoglu, J. Kitchin, T. F. Jaramillo, J. K. Nørskov, J. Rossmeisl, ChemCatChem, 2011, 3, 1159-1165.
[14]	O. Zhou, R. M. Fleming, D. W. Murphy, C. H. Chen, R. C. Haddon, A. P. Ramirez, S. H. Glarum, Science, 1994, 263, 1744-1747.
[15]	J. C. Charlier, Acc. Chem. Res. 2002, 35, 1063-1069.
[16]	Z. Hou, X. Wang, T. Ikeda, K. Terakura, M. Oshima, M.-A. Kakimoto, Phys. Rev. B, 2013, 87, 165401.
[17]	J. Kotakoski, A. V. Krasheninnikov, U. Kaiser, J. C. Meyer, Phys. Rev. Lett. 2011, 106, 105505.
[18]	A. Ruiz Puigdollers, P. Schlexer, S. Tosoni, G. Pacchioni, ACS Catal. 2017, 7, 6493-6513.
[19]	D. Hsu, IEEE Intell. Syst. 2016, 31, 60-60.
[20]	M. Wang, H. Zhu, ACS Catal. 2021, 11, 3930-3937.
[21]	L. Chanussot, A. Das, S. Goyal, T. Lavril, M. Shuaibi, M. Riviere, K. Tran, J. Heras-Domingo, C. Ho, W. Hu, A. Palizhati, A. Sriram, B. Wood, J. Yoon, D. Parikh, C. L. Zitnick, Z. Ulissi, ACS Catal. 2021, 11, 13062-13065.
[22]	S. Lu, Q. Zhou, Y. Ouyang, Y. Guo, Q. Li, J. Wang, Nat. Commun. 2018, 9, 3405.
[23]	B. Weng, Z. Song, R. Zhu, Q. Yan, Q. Sun, C. G. Grice, Y. Yan, W.-J. Yin, Nat. Commun. 2020, 11, 3513.
[24]	Z. C. Yu, W. M. Huang, Electroanalysis, 2022, 34, 599-607.
[25]	Z. Yang, W. Gao, Adv. Sci. 2022, 9, 2106043.
[26]	H. X. Mai, T. C. Le, D. H. Chen, D. A. Winkler, R. A. Caruso, Chem. Rev. 2022, 122, 13478-12515.
[27]	W. H. B. Sauer, M. K. Schwarz, J. Chem. Inf. Comput. Sci. 2003, 43, 987-1003.
[28]	J. Y. Qin, H. Liu, P. C. Zou, R. Zhang, C. Y. Wang, H. L. Xin, J. Am. Chem. Soc. 2022, 144, 2197-2207.
[29]	Y. Jia, X. Y. Xiong, D. N. Wang, X. X. Duan, K. Sun, Y. J. Li, L. R. Zheng, W. F. Lin, M. D. Done, G. X. Zhang, W. Liu, X. M. Sun, Nano-Micro. Lett. 2020, 12, 116.
[30]	H. J. Shen, E. Gracia-Espino, J. Y. Ma, H. D. Tang, X. Mamat, T. Wagberg, G. Z. Hu, S. J. Guo, Nano Energy, 2017, 35, 9-16.
[31]	X. L. Zhang, Z. X. Yang, Z. S. Lu, W. C. Wang, Carbon, 2018, 130, 112-119.
[32]	J. Zhang, Y. Wang, Y. Y. Wang, M. G. Zhang, J. Electrochem. Soc. 2017, 164, F1122-F1129.
[33]	F. F. Zhang, Y. L. Zhu, C. Tang, Y. Chen, B. B. Qian, Z. W. Hu, Y. C. Chang, C. W. Pao, Q. Lin, S. A. Kazemi, Y. Wang, L. Zhang, X. W. Zhang, H. T. Wang, Adv. Funct. Mater. 2022, 32, 2110224.
[34]	W. X. Chen, J. J. Pei, C. T. He, J. W. Wan, H. L. Ren, Y. Wang, J. C. Dong, K. L. Wu, W. C. Cheong, J. J. Mao, X. S. Zheng, W. S. Yan, Z. B. Zhuang, C. Chen, Q. Peng, D. S. Wang, Y. D. Li, Adv. Mater. 2018, 30, 1800396.
[35]	S. Back, J. Yoon, N. Tian, W. Zhong, K. Tran, Z. W. Ulissi, J. Phys. Chem. Lett. 2019, 10, 4401-4408.
[36]	T. Xie, J. C. Grossman, Phys. Rev. Lett. 2018, 120, 145301.
[37]	B. Bai, Y. Liu, C. Ma, G. Wang, G. Yan, K. Yan, M. Zhang, Z. Zhou, Sci. Sin. Math. 2020, 50, 367-384.
[38]	J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, G. E. Dahl, in: International Conference on Machine Learning, JMLR, Sydney, 2017, 1263-1272.
[39]	G. H. Gu, J. Noh, S. Kim, S. Back, Z. Ulissi, Y. Jung, J. Phys. Chem. Lett. 2020, 11, 3185-3191.
[40]	J. Wang, L. Kong, Z. Huang, J. Xiao, Comput. Eng. 2021, 47(4), 1-12.
[41]	X. Fang, L. Liu, J. Lei, D. He, S. Zhang, J. Zhou, F. Wang, H. Wu, H. Wang, Nat. Mach. Intell. 2022, 4, 127-134.
[42]	T. Long, N. M. Fortunato, I. Opahle, Y. Zhang, I. Samathrakis, C. Shen, O. Gutfleisch, H. Zhang, npj Comput. Mater. 2021, 7, 66.
[43]	J. Noh, J. Kim, H. S. Stein, B. Sanchez-Lengeling, J. M. Gregoire, A. Aspuru-Guzik, Y. Jung, Matter, 2019, 1, 1370-1384.
[44]	E. H. Cho, L.-C. Lin, J. Phys. Chem. Lett. 2021, 12, 2279-2285.
[45]	F. Gama, A. G. Marques, G. Leus, A. Ribeiro, in: Asilomar Conferenceon Signals, Systems, and Computers IEEE, Califonia, 2019, 452-456.
[46]	R. C. Gonzalez, IEEE Signal Process Mag. 2018, 35, 79-87.
[47]	F. Gama, E. Isufi, G. Leus, A. Ribeiro, IEEE Signal Process Mag. 2020, 37, 128-138.
[48]	C. Shorten, T. M. Khoshgoftaar, J. Big Data, 2019, 6, 60.
[49]	G. Kresse, J. Furthmuller, Comput. Mater. Sci. 1996, 6, 15-50.
[50]	G. Kresse, J. Furthmuller, Phys. Rev. B, 1996, 54, 11169-11686.
[51]	P. E. Blochl, Phys. Rev. B, 1994, 50, 17953-17979.
[52]	G. Kresse, D. Joubert, Phys. Rev. B, 1999, 59, 1758-1775.
[53]	J. P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 1996, 77, 3865-3868.
[54]	H. T. Chung, D. A. Cullen, D. Higgins, B. T. Sneed, E. F. Holby, K. L. More, P. Zelenay, Science, 2017, 357, 479-483.
[55]	J. Liu, M. G. Jiao, L. L. Lu, H. M. Barkholtz, Y. P. Li, Y. Wang, L. H. Jiang, Z. J. Wu, D. J. Liu, L. Zhuang, C. Ma, J. Zeng, B. S. Zhang, D. S. Su, P. Song, W. Xing, W. L. Xu, Y. Wang, Z. Jiang, G. Q. Sun, Nat. Commun. 2017, 8, 15938.
[56]	J. Liu, X. J. Sun, P. Song, Y. W. Zhang, W. Xing, W. L. Xu, Adv. Mater. 2013, 25, 6879-6883.
[57]	A. Zitolo, V. Goellner, V. Armel, M. T. Sougrati, T. Mineva, L. Stievano, E. Fonda, F. Jaouen, Nat. Mater. 2015, 14, 937-942.
[58]	M. T. M. Koper, J. Electroanal. Chem. 2011, 660, 254-260.
[59]	X. R. Zhu, J. X. Yan, M. Gu, T. Y. Liu, Y. F. Dai, Y. H. Gu, Y. F. Li, J. Phys. Chem. Lett. 2019, 10, 7760-7766.
[60]	H. Li, S. P. Xu, M. Wang, Z. H. Chen, F. F. Ji, K. W. Cheng, Z. Y. Gao, Z. Ding, W. J. Yang, J. Mater. Chem. A, 2020, 8, 17987-17997.
[61]	Y. M. Sun, H. B. Liao, J. R. Wang, B. Chen, S. N. Sun, S. J. H. Ong, S. B. Xi, C. Z. Diao, Y. H. Du, J. O. Wang, M. B. H. Breese, S. Z. Li, H. Zhang, Z. C. J. Xu, Nat. Catal. 2020, 3, 554-563.
[62]	X. H. Wan, Z. F. Zhang, H. Niu, Y. H. Yin, C. G. Kuai, J. Wang, C. Shao, Y. Z. Guo, J. Phys. Chem. Lett. 2021, 12, 6111-6118.
[63]	Y. Lecun, L. Bottou, Y. Bengio, P. Haffner, Proc. IEEE, 1998, 86, 2278-2324.
[64]	X. F. Liu, Y. F. Zhang, W. T. Wang, Y. Z. Chen, W. J. Xiao, T. Y. Liu, Z. Zhong, Z. J. Luo, Z. Ding, Z. F. Zhang, ACS Appl. Mater. Interfaces, 2022, 14, 1249-1259.
[65]	H. Niu, X. H. Wang, X. T. Wang, C. Shao, J. Robertson, Z. F. Zhang, Y. Z. Guo, ACS Sustainable Chem. Eng. 2021, 9, 3590-3599.
[66]	L. P. Wu, T. Guo, T. Li, iScience, 2021, 24, 102398.
[67]	S.-H. Wang, H. S. Pillai, S. Wang, L. E. K. Achenie, H. Xin, Nat. Commun. 2021, 12, 5288.
[68]	Z. Li, S. Wang, W. S. Chin, L. E. Achenie, H. Xin, J. Mater. Chem. A, 2017, 5, 24131-24138.
[69]	O. Mamun, K. T. Winther, J. R. Boes, T. Bligaard, npj Comput. Mater. 2020, 6, 177.
[70]	S. Wang, H. S. Pillai, H. Xin, Nat. Commun. 2020, 11, 6132.
[71]	V. Fung, G. Hu, P. Ganesh, B. G. Sumpter, Nat. Commun. 2021, 12, 88.
[72]	L. Breiman, Mach. Learn. 2001, 45, 5-32.
[73]	A. Fisher, C. Rudin, F. Dominici, J. Mach. Learn. Res. 2019, 20, 177.
[74]	R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Int. J. Comput. Vis., 2020, 128, 336-359.
[75]	L. van der Maaten, G. Hinton, J. Mach. Learn. Res. 2008, 9, 2579-2605.
[76]	R. Bro, A. K. Smilde, Anal. Methods, 2014, 6, 2812-2831.

用于高精度催化性能预测的精细结构敏感型深度学习框架

Fine-structure sensitive deep learning framework for predicting catalytic properties with high precision

RichHTML

PDF

Supporting Information

可视化

摘要/Abstract

引用本文

使用本文

扫码分享

图/表 11

参考文献

相关文章 15

编辑推荐

Metrics

[1]	孟帅奇, 李忠玉, 张鹏, Francisca Contreras, 季宇, Ulrich Schwaneberg. 深度学习指导的热裂褐藻角质酶工程用于促进PET解聚[J]. 催化学报, 2023, 50(7): 229-238.
[2]	夏思源, 李奇远, 张仕楠, 许冬, 林秀, 孙露菡, 许景三, 陈接胜, 李国栋, 李新昊. Pd纳米立方体异质结尺寸依赖的电子界面效应对苯酚加氢反应的促进作用[J]. 催化学报, 2023, 49(6): 180-187.
[3]	曾繁林, 朱虎林, 王茹楠, 袁晓亚, 孙凯, 屈凌波, 陈晓岚, 於兵. 一种用于光催化C(sp²)‒H键官能团化通用的非均相钒酸铋催化剂[J]. 催化学报, 2023, 46(3): 157-166.
[4]	王立晶, 杨天一, 冯博, 许祥雨, 申玉莹, 李孜涵, Arramel, 江吉周. 基于机器学习和第一性原理计算构建双电子转移通道加速CO₂光还原[J]. 催化学报, 2023, 54(11): 265-277.
[5]	李晓天, 陈林, 商城, 刘智攀. 调控炔烃半氢化反应的催化选择性: 实验和理论的最新进展[J]. 催化学报, 2022, 43(8): 1991-2000.
[6]	周紫璇, 高鹏. 多相催化CO₂加氢直接合成大宗化学品与液体燃料[J]. 催化学报, 2022, 43(8): 2045-2056.
[7]	陈乐添, 张旭, 陈安, 姚赛, 胡绪, 周震. 利用机器学习靶向设计先进电催化剂[J]. 催化学报, 2022, 43(1): 11-32.
[8]	严文娟, 张东培, 孙玉, 周子淇, 杜屹航, 杜一垚, 李玉姗, 刘梦媛, 张玉明, 沈健, 金鑫. 葡萄糖氧化制葡萄糖酸的催化剂结构敏感性研究进展[J]. 催化学报, 2020, 41(9): 1320-1336.
[9]	胡忠攀, 杨丹丹, 王政, 袁忠勇. 丙烷直接脱氢催化剂的研究进展[J]. 催化学报, 2019, 40(9): 1233-1254.
[10]	Koranteng Ernest, 刘以银, 刘思跃, 伍强贤, 陆良秋, 肖文精. 协同的非均相光氧化还原催化和镍催化构建碳磷键[J]. 催化学报, 2019, 40(12): 1841-1846.
[11]	李航, 吴博韬, 王建辉, 王富民, 张旭斌, 王刚, 李海朝. 一种高效稳定的低负载量的氯化-2-羟乙基三甲胺改性钌基催化剂用于乙炔氢氯化反应[J]. 催化学报, 2018, 39(11): 1770-1781.
[12]	梁锦霞, 杨小峰, 许聪俏, 张涛, 李隽. 单原子催化剂M1/FeO_x(M=Pt,Fe)的理论研究:CO氧化反应[J]. 催化学报, 2017, 38(9): 1566-1573.
[13]	焉晓明, 丁璇, 潘昱, 许小伟, 郝策, 郑文姬, 贺高红. 固载型季铵盐离子液体催化合成环状碳酸酯:烷基链长及羟基的影响[J]. 催化学报, 2017, 38(5): 862-871.
[14]	王伟, 徐振新, 郭章龙, 江成发, 储伟. 水滑石类催化剂催化制氢及生物炼制[J]. 催化学报, 2015, 36(2): 139-147.
[15]	尹冬菊, 张立忠, 赵秀峰, 陈汉, 翟倩. 铁-谷氨酸-硅钨酸三元配合物作为高活性非均相类Fenton催化剂降解4-氯酚[J]. 催化学报, 2015, 36(12): 2203-2210.