| 182 | 0 | 68 |
| 下载次数 | 被引频次 | 阅读次数 |
有机化学正向反应预测作为人工智能在化学信息学领域的重要应用方向，一直受到研究者的关注。简化分子线性输入系统（simplified molecular-input line-entry system,SMILES）为化学分子结构提供了一种标准化的线性编码方法。基于此，可以将正向反应预测任务转换为机器翻译中序列到序列的生成任务。传统的Transformer模型只关注化学式中原子间的注意力权重，忽略分子内的全局信息；此外，传统的Transformer模型面对同一个分子的不同表示的时候，存在着预测性能下降、泛化能力不强的问题。为了解决上述问题，本文提出一种基于双编码器的多层级有机化学的反应预测方法。首先，采用2个编码器分别处理分子层级和原子层级的信息，使得模型专注于各自层面的特征提取；其次，提出分子特征表示算法，对原子嵌入取平均获得分子间的相互作用；最后，利用自适应的门控单元对原子编码器和分子编码器的输出结果进行多层级特征融合，并将融合结果输入到解码器。实验结果表明，所提出的方法在3个USPTO数据集上的性能优于基线模型，并且提升了模型的泛化能力和长序列的处理能力。
Abstract：Prediction of forward reactions in organic chemistry as a critical artificial intelligence(AI) application has attracted more attention from researchers in recent years. The simplified molecular-input line-entry system(SMILES)provides a method to linearize the chemical molecular formula. Thus, based on SMILES encoding, reaction prediction can be converted to the task of sequence-to-sequence generation in a neural machine translation(NMT). The traditional Transformer model focuses only on the inter-atomic attention weights in the chemical formula and ignores the global information within the molecule. Moreover, the traditional Transformers suffer from performance degradation and limited generalization when encountering diverse representations of identical molecules. To address these points,we propose a reaction prediction method for multilevel organic chemistry based on dual encoders. Firstly, our model uses two encoders to process the information at the molecular level and the atomic level, respectively. Next, a molecular feature algorithm is proposed to obtain the interaction between molecules by averaging atom embeddings. Finally, a gating unit for automatic adjustment is utilized to perform multi-level feature fusion between the outputs of the atomic and molecular encoders. The fusion result is input to the decoder. Our results illustrate the method proposed obtains better results than the baseline on the three USPTO datasets and improves the generalization ability and long sequence processing ability of the model.
[1]COREY E J, WIPKE W T. Computer-assisted design of complex organic syntheses[J]. Science, 1969, 166(3902):178-192.
[2]SATOH H, FUNATSU K. Further development of a reaction generator in the SOPHIA system for organic reaction prediction. Knowledge-guided addition of suitable atoms and/or atomic groups to product skeleton[J]. Journal of Chemical Information and Computer Sciences, 1996, 36(2):173-184.
[3]COLEY C W , BARZILAY R , JAAKKOLA T S , et al.Prediction of organic reaction outcomes using machine learning[J]. ACS Central Science, 2017, 3(5):434-443.
[4]DUVENAUD D , MACLAURIN D , AGUILERA-IPARRAGUIRRE J, et al. Convolutional networks on graphs for learning molecular fingerprints[C]//Proceedings of the 29th International Conference on Neural Information Processing Systems-Volume 2, December 7-12, 2015, Montreal,Canada. Cambridge, MA:MIT Press, 2015:2224-2232.
[5]RACC UGLIA P, ELBERT K C, ADLER P D F, et al.Machine-learning-assisted materials discovery using failed experiments[J]. Nature, 2016, 533(7601):73-76.
[6]SEGLER M H S, WALLER M P. Modelling chemical reasoning to predict and invent reactions[J]. Chemistry-A European Journal, 2017, 23(25):6118-6128.
[7]WEININGER D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules[J]. Journal of Chemical Information and Computer Sciences, 1988, 28(1):31-36.
[8]CASEY R G, LECOLINET E. A survey of methods and strategies in character segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1996, 18(7):690-706.
[9]ZHANG Z S, ZHAO H, LING K W, et al. Effective subword segmentation for text comprehension[J]. IEEE/ACM Transactions on Audio , Speech , and Language Processing, 2019, 27(11):1664-1674.
[10]GALL魪M. Investigating the effectiveness of BPE:the power of shorter sequences[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing(EMNLP-IJCNLP), Hong Kong,China. Stroudsburg, PA, USA:Association for Computational Linguistics, 2019:1375-1381.
[11]SCHWALLER P, LAINO T, GAUDIN T, et al. Molecular transformer:a model for uncertainty-calibrated chemical reaction prediction[J]. ACS Central Science, 2019, 5(9):1572-1583.
[12]LO A, POLLICE R, NIGAM A, et al. Recent advances in the self-referencing embedded strings(SELFIES)library[J].Digital Discovery, 2023, 2(4):897-908.
[13]UCAK U V, ASHYRMAMATOV I, LEE J. Improving the quality of chemical language model outcomes with atomin-SMILES tokenization[J]. Journal of Cheminformatics ,2023, 15:55.
[14]LU J Y, ZHANG Y K. Unified deep learn ing model for multitask reaction predictions with explanation[J]. Journal of Chemical Information and Modeling, 2022, 62(6):1376-1387.
[15]TANG G B, MÜLLER M, RIOS A, et al. Why self-attention? A targeted evaluation of neural machine translation architectures[EB/OL].(2018-10-27)[2024-09-27]. https://arxiv.org/abs/1808.08946.
[16]WU F, FAN A, BAEVSKI A, et al. Pay less attention with lightweight and dynamic convolutions[EB/OL].(2019-01-29)[2024-09-27]. https://arxiv.org/abs/1901.10430.
[17]VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[EB/OL].(2017-06-12)[2024-09-27].https://arxiv.org/abs/1706.03762.
[18]GOTHARD C M, SOH S, GOTHARD N A, et al. Rewiring chemistry:algorithmic discovery and experimental validation of one-pot reactions in the network of organic chemistry[J]. Angewandte Chemie International Edition, 2012,51(32):7922-7927.
[19]DUIGOU T , DU LAC M , CARBONELL P , et al.RetroRules:a database of reaction rules for engineering biology[J]. Nucleic Acids Research, 2019, 47(D1):D1229-D1235.
[20]LIN K J, XU Y J, PEI J F, et al. Automatic retrosynthetic route planning using template-free models[J]. Chemical Science, 2020, 11(12):3355-3364.
[21]WANG L P, MCGIBBON R T, PANDE V S, et al. Automated discovery and refinement of reactive molecular dynamics pathways[J]. Journal of Chemical Theory and Computation, 2016, 12(2):638-649.
[22]MAEDA S, HARABUCHI Y, HAYASHI H, et al. Toward ab initio reaction discovery using the artificial force induced reaction method[J]. Annual Review of Physical Chemistry, 2023, 74:287-311.
[23]WU Z X, JIANG D J, WANG J K, et al. Knowledgebased BERT:a method to extract molecular features like computational chemists[J]. Briefings in Bioinformatics ,2022, 23(3):bbac131.
[24]CHEN S, JUNG Y. Deep retrosynthetic reaction prediction using local reactivity and global attention[J]. JACS Au ,2021, 1(10):1612-1620.
[25]LIU Z Q, ZHANG W, XIA Y C, et al. MolXPT:wrapping molecules with text for generative pre-training[EB/OL].(2023-05-18)[2024-09-27]. https://arxiv.org/abs/2305.10688.
[26]JIN D, JIN Z J, ZHOU J T, et al. Is BERT really robust?A strong baseline for natural language attack on text classification and entailment[J]. Proceedings of the AAAI Conference on Artificial Intelli gence , 2020 , 34(5):8018-8025.
[27]WANG T T. Research on chemical reaction prediction model based on Fairseq[C]//Proceedings of the 2021 2nd International Conference on Big Data&Artificial Intelligence&Software Engineering(ICBASE), September 24-26, 2021, Zhuhai, China. New York:IEEE Xplore, 2021:167-171.
[28]MICULICICH L, RAM D, PAPPAS N, et al. Documentlevel neural machine translation with hierarchical attention networks[EB/OL].(2018-09-05)[2024-09-27]. https://arxiv.org/abs/1809.01576.
[29]PAPINENI K , ROUKOS S, WARD T, et al. BLEU:a method for automatic evaluation of machine translation[C]//Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics. Philadelphia, Pennsylvania,USA:Association for Computational Linguistics, 2002:311-318.
[30]JIN W G, COLEY C W, BARZILAY R, et al. Predicting organic reaction outcomes with Weisfeiler-Lehman network[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, December 4-9,2017, Long Beach, California, USA. New York:Curran Associates Inc., 2017:2604-2613.
[31]SCHWALLER P, GAUDIN T, LÁNYI D, et al.″Found in translation″：predicting outcomes of complex organic chemistry reactions using neural sequence-to-sequence models[J]. Chemical Science, 2018, 9(28):6091-6098.
[32]BJERRUM E J. SMILES enumeration as data augmentation for neural network modeling of molecules[EB/OL].(2017-03-21)[2024-09-27]. https://arxiv.org/abs/1703.07076.
[33]TETKO I V, KARPOV P, van DEURSEN R, et al. Stateof-the-art augmented NLP transformer models for direct and single-step retrosynthesis[J]. Nature Communications,2020, 11:5575.
[34]KOOLSTRA K, B魻RNERT P, LELIEVELDT B P F, et al.Stochastic neighbor embedding as a tool for visualizing the encoding capability of magnetic resonance fingerprinting dictionaries[J]. Magnetic Resonance Materials in Physics,Biology and Medicine, 2022, 35(2):223-234.
[35]AL KHALIFA A, HARANCZYK M, HOLLIDAY J. Comparison of nonbinary similarity coefficients for similarity searching, clustering and compound selection[J]. Journal of Chemical Information and Modeling, 2009, 49(5):1193-1201.
基本信息:
DOI：10.12194/j.ntu.20241127001
中图分类号:TP18;TQ203
引用信息:
[1]朱林星,姜舒,黄嘉爽,等.基于双编码器的多层级有机化学的反应预测[J].南通大学学报(自然科学版),2026,25(02):24-35+70.DOI:10.12194/j.ntu.20241127001.
基金信息:
国家自然科学基金面上项目(62471259);国家自然科学基金青年科学基金项目(62406153)
2025-05-14
2025-05-14
2025-05-14