基于反向机器学习的调香设计方法

doi:10.11949/0438-1157.20191015

摘要/Abstract

摘要：

香精香料业务拥有数十亿美元的市场，而调香技术的发展丰富了人们的现代社会生活。提出了用于调香设计的反向机器学习模型，采用基于类导体屏蔽模型（COSMO）得到的分子表面电荷密度分布作为香精分子的结构描述符设计最终的调香产品。首先，识别香精属性并根据需求将属性转化为目标性质。然后，调节气味并建立了反向机器学习（inverse machine learning, IML）模型，其中输入变量为气味，输出变量为分子结构描述符，利用训练得到的IML模型，根据目标特性，预测出潜在产品的结构描述符。最后在指定的数据库中通过欧几里得距离法筛选出候选的调香混合物。本文还以两类调香算例为例，利用该框架进行了调香设计，并利用实验数据及气味雷达图对实验结果进行了验证。

关键词: 系统工程, 设计, 模型, 反向机器学习, 调香, 分子表面电荷密度分布

Abstract:

The business of fragrances has become a multibillion-dollar market, and the development of fragrance tuned technology enriches modern social life. In this study, the inverse machine learning model for fragrance tuned design is proposed. The molecular surface charge density distribution based on the conductor-like screening model (COSMO) is used as the structural descriptor of the fragrance molecule to design the final fragrance tuned product. First, the fragrance attributes are identified and transform attributes into target properties according to needs. Then, change odor scores and establish the Inverse Machine Learning (IML) models, in which the input variables are odors and the output variable is molecular structure descriptor. Based on the trained IML models, the structure descriptors of the potential product are predicted according to the target properties. Finally, the candidate tuned mixtures were screened out using Euclidean-based method in the specified database. In this paper, two types of fragrant examples are taken as examples. The framework is used to design the fragrance, and the experimental data and odor radar map are used to verify the experimental results.

Key words: systems engineering, design, model, inverse machine learning, tuned fragrance, molecular surface charge density distribution

中图分类号:

TQ 021.8

王璐, 毛海涛, 张磊, 刘琳琳, 都健. 基于反向机器学习的调香设计方法[J]. 化工学报, 2019, 70(12): 4722-4729.

Lu WANG, Haitao MAO, Lei ZHANG, Linlin LIU, Jian DU. Inverse machine learning-based fragrance tuned design method[J]. CIESC Journal, 2019, 70(12): 4722-4729.

图/表 7

参考文献 30

1	Shibamoto T , Mihara S . Photochemistry of fragrance materials(Ⅰ): Unsaturated compounds[J]. Journal of Toxicology Cutaneous & Ocular Toxicology, 1983, 2(2/3): 153-192.
2	王洪记 . 国际食用香精市场分析[J]. 精细与专用化学品, 1997, 17: 16.
	Wang H J . International fragrance edible market analysis[J]. Fine and Specialty Chemicals, 1997, 17: 16.
3	胡勇成 . 调香方法的一些认识[C]// 2006年中国香料香精学术研讨会. 上海， 2006: 248-254.
	Hu Y C . Development and Issues on Compoundings[C]// Proceedings of the Symposium on flavor and fragrance in China. Shanghai, 2006: 248-254.
4	Holmes J C , Morrell F A . Oscillographic mass spectrometric monitoring of gas chromatography[J]. Applied Spectroscopy, 1957, 11(2): 86-87.
5	NIST . National Institute of Standards and Technology[M]. America: National Institute of Standards and Technology, 1988.
6	林旭辉, 刘平, 李楠 . 食品香精香料及加香技术[M]. 北京: 中国轻工业出版社, 2010: 149-157.
	Lin X H , Liu P , Li N . Food Flavors and Fragrances and Flavoring Techniques[M]. Beijing: China Light Industry Press, 2010: 149-157.
7	孙宝国 . 肉味香精技术进展[J]. 食品科学, 2004, 25(10): 339-342.
	Sun B G . The development of meat flavor technology[J]. Food Science, 2004, 25(10): 339-342.
8	Hornic K . Multilayer feedforward networks are universal approximators[J]. Neural Networks, 1989, 2: 359-366.
9	Carrera G V S M , Gupta S , Aires-De-Sousa J . Machine learning of chemical reactivity from databases of organic reactions[J]. J. Comput. Aided Mol. Des., 2009, 23(7): 419-429.
10	Venkatraman V , Evjen S , Knuutila H K , et al . Predicting ionic liquid melting points using machine learning[J]. Journal of Molecular Liquids, 2018, 264: 318-326.
11	Li Z , Ma X , Xin H . Feature engineering of machine-learning chemisorption models for catalyst design[J]. Catalysis Today, 2016, 280(2): 232-238.
12	Spellings M , Glotzer S C . Machine learning for crystal identification and discovery[J]. AIChE Journal, 2018, 64(6): 2198-2206.
13	Zhang L , Mao H , Liu L , et al . A machine learning based computer-aided molecular design/screening methodology for fragrance molecules[J]. Computers & Chemical Engineering, 2018, 115: 295-308.
14	Holzinger A . Data mining with decision trees: theory and applications[J]. Online Information Review, 2015, 39(3): 437-438.
15	Velásco-Mejía A , Vallejo-Becerra V , Chávez-Ramírez A U , et al . Modeling and optimization of a pharmaceutical crystallization process by using neural networks and genetic algorithms[J]. Powder Technology, 2016, 292(11): 122-128.
16	Toubaei A , Golmohammadi H , Dashtbozorgi Z , et al . QSPR studies for predicting gas to acetone and gas to acetonitrile solvation enthalpies using support vector machine[J]. Journal of Molecular Liquids, 2012, 175: 24-32.
17	Zonouz P R , Niaei A , Tarjomannejad A . Modeling and optimization of toluene oxidation over perovskite-type nanocatalysts using a hybrid artificial neural network-genetic algorithm method[J]. Journal of the Taiwan Institute of Chemical Engineers, 2016, 65: 276-285.
18	Keller A , Gerkin R C , Guan Y , et al . Predicting human olfactory perception from chemical features of odor molecules[J]. Science, 2017, 355(6327): 820-826.
19	Keller A , Vosshall L B . Olfactory perception of chemically diverse molecules[J]. BMC Neuroscience, 2016, 17(1): 55.
20	Gani R , Brignole E A . Molecular design of solvents for liquid extraction based on UNIFAC[J]. Fluid Phase Equilibria, 1983, 13(83): 331-340.
21	Ajmani S , Jadhav K , Kulkarni S A . Group-based QSAR (G-QSAR): mitigating interpretation challenges in QSAR[J]. Molecular Informatics, 2010, 28(1): 36-51.
22	Klamt A , Schüürmann G . COSMO: a new approach to dielectric screening in solvents with explicit expressions for the screening energy and its gradient[J]. J. Chem. Soc. Perkin Transactions, 1993, 2(5): 799-805.
23	Teixeira M A , Rodríguez O , Gomes P , et al . Chapter 1 - A Product Engineering Approach in the Perfume Industry[M]. England: Perfume Engineering, 2013: 1-13.
24	Nie L , Sun J , Huang R . The biosynthesis and affecting factors of aroma in some fruits[J]. Chinese Bulletin of Botany, 2004, 21(5): 631-637.
25	Lavine B K , Davidson C E , Breneman C , et al . Electronic van der Waals surface property descriptors and genetic algorithms for developing structure-activity correlations in olfactory databases[J]. Journal of Chemical Information & Computer Sciences, 2003, 43(6): 1890.
26	Lin S T , Sandler S I . A priori phase equilibrium prediction from a segment contribution solvation model[J]. Industrial & Engineering Chemistry Research, 2002, 41(5): 899-913.
27	Palomar J , Torrecilla J S , Ferro V R , et al . Development of an a priori ionic liquid design tool(1): Integration of a novel COSMO-RS molecular descriptor on neural networks[J]. Industrial & Engineering Chemistry Research, 2008, 47(13): 4523-4532.
28	Hukkerikar A S . Development of pure component property models for chemical product-process design and analysis[D]. Denmark: Technical University of Denmark, 2013.
29	Reid R C , Prausnitz J M , Poling B E . The Properties of Gases & Liquids[M]. New York: McGraw Hill, 1988.
30	Miller F P , Vandome A F , Mcbrewster J . Coefficient of determination[J]. Alphascript Publishing, 2006, 31(1): 63-64.

模型	R ²
模型	S1	S2	S3	S4	S5	S6	S7	S8	S9	S10	平均值
ANN	0.9414	0.8312	0.8578	0.8546	0.8841	0.8781	0.9378	0.8576	0.8661	0.8979	0.8807
SVM	0.00004	0.3864	0.0341	0.0050	0.8381	0.8588	0.3997	0.0123	0.0638	0.0000	0.2598
RF	0.0435	0.2041	0.4981	0.1931	0.6279	0.6386	0.1663	0.0557	0.5902	0.0000	0.3018
MLR	0.0275	0.0879	0.0125	0.0505	0.8760	0.8683	0.3629	0.2398	0.0475	0.0000	0.2573

模型	R ²
模型	S1	S2	S3	S4	S5	S6	S7	S8	S9	S10	平均值
ANN	0.9414	0.8312	0.8578	0.8546	0.8841	0.8781	0.9378	0.8576	0.8661	0.8979	0.8807
SVM	0.00004	0.3864	0.0341	0.0050	0.8381	0.8588	0.3997	0.0123	0.0638	0.0000	0.2598
RF	0.0435	0.2041	0.4981	0.1931	0.6279	0.6386	0.1663	0.0557	0.5902	0.0000	0.3018
MLR	0.0275	0.0879	0.0125	0.0505	0.8760	0.8683	0.3629	0.2398	0.0475	0.0000	0.2573

项目	4-庚烯醛（CAS No.:6728-31-0；FEMA:3289）		惕各酸烯丙酯（CAS No.:7493-71-2；FEMA:2043）
项目	原始	目标	原始	目标
愉悦度	29.2	29.2	42.13	42.13
可食用味	25.59	35.00↑	32	32
烘焙味	38.11	38.11	30.8	30.8
甜味	12.46	30.00↑	33.85	33.85
水果味	11.55	25.00↑	29.5	29.5
馊味	26.83	26.83	41.94	30.00↓
花香味	24.33	24.33	33.46	33.46

项目	4-庚烯醛（CAS No.:6728-31-0；FEMA:3289）		惕各酸烯丙酯（CAS No.:7493-71-2；FEMA:2043）
项目	原始	目标	原始	目标
愉悦度	29.2	29.2	42.13	42.13
可食用味	25.59	35.00↑	32	32
烘焙味	38.11	38.11	30.8	30.8
甜味	12.46	30.00↑	33.85	33.85
水果味	11.55	25.00↑	29.5	29.5
馊味	26.83	26.83	41.94	30.00↓
花香味	24.33	24.33	33.46	33.46

性质	4-庚烯醛	惕各酸烯丙酯
溶解度^①(298 K, 水)/(mg·L^-1)	1810	573.6
沸点^①(101.325 kPa)/K	429.40	445.57
K _o/w ^①	2.174	2.075
LC ₅₀ ^②/(mol·L^-1)	3.68	4.97
闪点^②/K	316.33	333