基于图注意力堆叠自编码器微生物-药物关联预测

王波; 何洋; 杜晓昕; 张剑飞; 徐靖然; 贾娜

doi:10.13700/j.bh.1001-5965.2023.0730

基于图注意力堆叠自编码器微生物-药物关联预测

doi: 10.13700/j.bh.1001-5965.2023.0730

王波^{1, 2, ,},
何洋¹,
杜晓昕^{1, 2},
张剑飞^{1, 2},
徐靖然¹,
贾娜¹

1.
齐齐哈尔大学计算机与控制工程学院，齐齐哈尔 161006
2.
齐齐哈尔大学黑龙江省大数据网络安全检测分析重点实验室，齐齐哈尔 161006

基金项目:

黑龙江省省属高等学校基本科研业务费国自然培育一般项目(145409324)

详细信息

通讯作者:
E-mail：bowangdr@qqhru.edu.cn

中图分类号: TP391
计量
- 文章访问数: 716
- HTML全文浏览量: 212
- PDF下载量: 54
- 被引次数: 0
出版历程
- 收稿日期: 2023-11-07
- 录用日期: 2024-01-12
- 网络出版日期: 2024-03-09
- 整期出版日期: 2026-01-15

Prediction of microbe-drug association based on graph attention stacked autoencoder

WANG Bo^{1, 2
, ,},
HE Yang¹,
DU Xiaoxin^{1, 2},
ZHANG Jianfei^{1, 2},
XU Jingran¹,
JIA Na¹

1.
College of Computer and Control Engineering，Qiqihar University，Qiqihar 161006，China
2.
Heilongjiang Key Laboratory of Big Data Network Security Detection and Analysis，Qiqihar University，Qiqihar 161006，China

Funds:

General Program for National Natural Science Foundation Cultivation under the Basic Research Fund for Provincial Universities in Heilongjiang (145409324)

More Information

Corresponding author: E-mail：bowangdr@qqhru.edu.cn

摘要

摘要:
传统方法发掘微生物与药物新关联主要通过生物实验完成，耗费时间且开销极大。基于此，提出基于图注意力堆叠自编码器微生物与药物关联预测方法GATSAE。建立微生物与药物异构网络，丰富关联信息；通过图卷积网络(GCN)提取多层潜在特征，得到微生物和药物的卷积融合矩阵；采用改进的堆叠自编码器学习有意义的高阶相似特征的无监督低维表示，在堆叠自编码器的基础上追加图卷积和注意力机制，进一步优化高阶特征信息的提取；将低维特征与关联特征串联，使用多层感知机(MLP)对最终的微生物-药物进行评分预测。通过效能评估，GATSAE方法的受试者工作特征曲线下面积(AUROC)及精确率-召回率曲线下面积(AUPR)分别达到0.9619和0.9577，优于经典的机器学习方法和常见的深度学习方法。案例研究表明，GATSAE方法能够准确预测到与SARS-CoV-2、大肠杆菌相关的候选药物，以及与阿司匹林相关的候选微生物。
- 微生物与药物 /
- 关联预测 /
- 堆叠自编码器 /
- 注意力机制 /
- 图卷积网络 /
- 多层感知机
Abstract:
A graph attention stacking autoencoder approach for predicting the association between microorganisms and drugs, known as GATSAE, is proposed in response to the conventional method of finding new associations between microorganisms and drugs, which is primarily accomplished through biological experiments, which is highly costly and time-consuming. Firstly, establish a heterogeneous network of microorganisms and drugs to enrich the associated information. Secondly, the convolutional fusion matrix of microorganisms and drugs is obtained by extracting multi-layer latent features through graph convolutional network (GCN). Once again, an improved stacked autoencoder is used to learn unsupervised low dimensional representations of meaningful high-order similar features. Graph convolution and attention mechanisms are added to the stacked autoencoder to further optimize the extraction of high-order feature information. Finally, the low-dimensional features are concatenated with associated features, and a multi-layer perceptron (MLP) is used to score and predict the final microbial drug. According to performance evaluation, GATSAE subjects’ area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPR) were 0.9619 and 0.9577, respectively. These results are better than those of popular deep learning techniques and traditional machine learning techniques. Case studies have shown that GATSAE can accurately predict candidate drugs related to SARS-CoV-2 and Escherichia coli, as well as candidate microorganisms related to aspirin.
- microbes and drugs /
- correlation prediction /
- stacked autoencoder /
- attention mechanism /
- graph convolutional network /
- multi-layer perceptron

HTML全文

图 1 GATSAE模型框架

Figure 1. Framework of GATSAE model

下载: 全尺寸图片幻灯片

图 2 图注意力堆叠自编码器结构

Figure 2. Structure of graph attention stacking autoencoder

下载: 全尺寸图片幻灯片

图 3 GATSAE交叉验证下的ROC曲线和PR曲线

Figure 3. ROC and PR curves under GATSAE cross validation

下载: 全尺寸图片幻灯片

图 4 不同卷积核带宽γ对应模型AUPR值

Figure 4. Bandwidth of different convolutional kernels γ Corresponding model AUPR value

下载: 全尺寸图片幻灯片

图 5 不同卷积层层数l对非卷积模型AUPR值的增长率

Figure 5. Growth rate of AUPR values in non convolutional models with different number of convolutional layers l

下载: 全尺寸图片幻灯片

图 6 不同学习率对训练损失的变化值

Figure 6. Variation value of training loss under different learning rates

下载: 全尺寸图片幻灯片

图 7 消融实验下的ROC和PR曲线对比

Figure 7. Comparison of ROC and PR curves under ablation experiments

下载: 全尺寸图片幻灯片

图 8 不同数据集下GATSAE预测结果对比

Figure 8. Comparison of GATSAE prediction results under different datasets

下载: 全尺寸图片幻灯片

图 9 MLP与常见分类模型比较

Figure 9. Comparison between MLP and common classification models

下载: 全尺寸图片幻灯片

图 10 模型指标对比

Figure 10. Comparison of model indicators

下载: 全尺寸图片幻灯片

表 1 3种数据集的数据记录

Table 1. Data recording for three datasets

数据集	药物种类	微生物种类	已知关联关系种类
MDAD	627	142	1152
aBiofilm	1720	140	2884
DrugVirus	175	95	933

下载: 导出CSV

表 2 对比实验分组信息

Table 2. Comparison experiment grouping information

实验组	GCN	图注意力堆叠自编码器	数据串联
分组1	引入	无	无
分组2	无	引入	无
分组3	无	无	引入
分组4	无	无	无
GATSAE	引入	引入	引入

下载: 导出CSV

表 3 MLP与常见分类模型评价指标

Table 3. Evaluation indicators of MLP and common classification models

分类器	AUC	AUPR	P_re	R_ec	F₁
DT	0.9163	0.9164	0.9126	0.9203	0.9162
RF	0.9366	0.9309	0.9124	0.8915	0.9214
KNN	0.9489	0.9486	0.9041	0.9161	0.9098
SVM	0.8744	0.8469	0.8212	0.8516	0.8357
MLP	0.9619	0.9577	0.9166	0.9500	0.9329

下载: 导出CSV

表 4 模型指标对比值

Table 4. Comparison value of model indicators

模型	AUC	AUPR	P_re	R_ec	F₁
GCNMDA	0.9240	0.9124	0.8817	0.9257	0.9029
EGATMDA	0.9434	0.9335	0.9015	0.9213	0.9118
HGATDVA	0.9159	0.9008	0.8976	0.8989	0.8971
HNERMDA	0.8977	0.9026	0.8537	0.8430	0.8463
NIRBMMDA	0.8485	0.8327	0.8391	0.8388	0.8268
GATSAE	0.9619	0.9577	0.9166	0.9300	0.9329

下载: 导出CSV

表 5 与大肠杆菌相关排名前10的药物

Table 5. Top 10 drugs related to Escherichia coli

排名	相关药物	PMID
1	Ceftizoxime	6299968
2	Aminosalicylic Acid	33468700
3	Citral	35776056
4	Clozapine	25448498
5	Palmitic acid	29719215
6	Esculin	15137927
7	Azlocillin	7033199
8	Azidocillin	4563142
9	Aspirin	30658983
10	Glipizide	32995125

下载: 导出CSV

表 6 与阿司匹林相关排名前10的微生物

Table 6. Top 10 microbes related to Aspirin

排名	相关微生物	PMID
1	Candida albicans	33242673
2	Pseudomonas aeruginosa	25088031
3	Staphylococcus epidermidis	12555346
4	Human immunodeficiency virus	28480270
5	Streptococcus mutans	unconfirmed
6	Staphylococcus aureus	34692677
7	Mycobacterium tuberculosis	23997233
8	Escherichia coli	30658983
9	Clostridium perfringens	31865684
10	Human herpesvirus	unconfirmed

下载: 导出CSV

表 7 与SARS-CoV-2相关排名前20的药物

Table 7. Top 20 drugs related to SARS-CoV-2

排名	相关微生物	PMID
1	Chloroquine	35859449
2	ABT	37414987
3	Favipiravir	33108587
4	BCX	35062212
5	Luteolin	32389723
6	Amodiaquine	32486229
7	Cyclosporine	34081806
8	Emetine	33302852
9	Gemcitabine	32432977
10	Hydroxychloroquine	32373993
11	Amiodarone	36426888
12	Obatoclax	34989664
13	Remdesivir	33436624
14	Chlorpromazine	32773341
15	Nelfinavir	35390430
16	EIPA	37632140
17	Arbidol	32955901
18	Niclosamide	35348204
19	Dasatinib	36704839
20	Eflornithine	34055746

下载: 导出CSV

参考文献(41)

[1]	杨博图. 基于相似性信息的微生物-药物关联关系预测方法研究[D]. 长沙: 中南大学, 2022: 69. YANG B T. Study on prediction method of microbial-drug correlation based on similarity information[D]. Changsha: Central South University, 2022: 69(in Chinese).
[2]	SHREINER A B, KAO J Y, YOUNG V B. The gut microbiome in health and in disease[J]. Current Opinion in Gastroenterology, 2015, 31(1): 69-75.
[3]	LEY R E, BÄCKHED F, TURNBAUGH P, et al. Obesity alters gut microbial ecology[J]. Proceedings of the National Academy of Sciences of the United States of America, 2005, 102(31): 11070-11075.
[4]	TURNBAUGH P J, RIDAURA V K, FAITH J J, et al. The effect of diet on the human gut microbiome: a metagenomic analysis in humanized gnotobiotic mice[J]. Science Translational Medicine, 2009, 1(6): 6ra14.
[5]	GOLDMAN E. Antibiotic abuse in animal agriculture: exacerbating drug resistance in human pathogens[J]. Human and Ecological Risk Assessment: an International Journal, 2004, 10(1): 121-134.
[6]	VRBANAC A, DEBELIUS J W, JIANG L J, et al. An elegan(t) screen for drug-microbe interactions[J]. Cell Host & Microbe, 2017, 21(5): 555-556.
[7]	AARNOUDSE A L H J, DIELEMAN J P, VISSER L E, et al. Common ATP-binding cassette B1 variants are associated with increased digoxin serum concentration[J]. Pharmacogenetics and Genomics, 2008, 18(4): 299-305.
[8]	HAISER H J, SEIM K L, BALSKUS E P, et al. Mechanistic insight into digoxin inactivation by Eggerthella lenta augments our understanding of its pharmacokinetics[J]. Gut Microbes, 2014, 5(2): 233-238.
[9]	ONG F S, DEIGNAN J L, KUO J Z, et al. Clinical utility of pharmacogenetic biomarkers in cardiovascular therapeutics: a challenge for clinical implementation[J]. Pharmacogenomics, 2012, 13(4): 465-475.
[10]	VOORA D, SHAH S H, SPASOJEVIC I, et al. The SLCO1B1*5 genetic variant is associated with statin-induced side effects[J]. Journal of the American College of Cardiology, 2009, 54(17): 1609-1616.
[11]	RAMSEY L B, JOHNSON S G, CAUDLE K E, et al. The clinical pharmacogenetics implementation consortium guideline for SLCO1B1 and simvastatin-induced myopathy: 2014 update[J]. Clinical Pharmacology & Therapeutics, 2014, 96(4): 423-428.
[12]	VIOLI F, LIP G Y, PIGNATELLI P, et al. Interaction between dietary vitamin K intake and anticoagulation by vitamin K antagonists: is it really true? : a systematic review[J]. Medicine, 2016, 95(10): e2895.
[13]	GUTHRIE L, GUPTA S, DAILY J, et al. Human microbiome signatures of differential colorectal cancer drug metabolism[J]. NPJ Biofilms and Microbiomes, 2017, 3: 27.
[14]	ZHU L Z, DUAN G H, YAN C, et al. Prediction of microbe-drug associations based on chemical structures and the KATZ measure[J]. Current Bioinformatics, 2021, 16(6): 807-819.
[15]	MA Y J, LIU Q Q. Generalized matrix factorization based on weighted hypergraph learning for microbe-drug association prediction[J]. Computers in Biology and Medicine, 2022, 145: 105503.
[16]	ZHU B, XU Y, ZHAO P C, et al. NNAN: nearest neighbor attention network to predict drug-microbe associations[J]. Frontiers in Microbiology, 2022, 13: 846915.
[17]	LONG Y H, WU M, KWOH C K, et al. Predicting human microbe-drug associations via graph convolutional network with conditional random field[J]. Bioinformatics, 2020, 36(19): 4918-4927.
[18]	LONG Y H, WU M, LIU Y, et al. Ensembling graph attention networks for human microbe-drug association prediction[J]. Bioinformatics, 2020, 36(Supplement_2): i779-i786.
[19]	LONG Y H, ZHANG Y, WU M, et al. Predicting drugs for COVID-19/SARS-CoV-2 via heterogeneous graph attention networks[C]// Proceedings of the IEEE International Conference on Bioinformatics and Biomedicine. Piscataway: IEEE Press, 2020: 455-459.
[20]	SUN Y Z, ZHANG D H, CAI S B, et al. MDAD: a special resource for microbe-drug associations[J]. Frontiers in Cellular and Infection Microbiology, 2018, 8: 424.
[21]	RAJPUT A, THAKUR A, SHARMA S, et al. aBiofilm: a resource of anti-biofilm agents and their potential implications in targeting antibiotic drug resistance[J]. Nucleic Acids Research, 2018, 46(D1): D894-D900.
[22]	ANDERSEN P I, IANEVSKI A, LYSVAND H, et al. Discovery and development of safe-in-man broad-spectrum antiviral agents[J]. International Journal of Infectious Diseases, 2020, 93: 268-276.
[23]	STEINBECK C, HOPPE C, KUHN S, et al. Recent developments of the chemistry development kit (CDK): an open-source Java library for chemo- and bioinformatics[J]. Current Pharmaceutical Design, 2006, 12(17): 2111-2120.
[24]	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.
[25]	KAMNEVA O K. Genome composition and phylogeny of microbes predict their co-occurrence in the environment[J]. PLoS Computational Biology, 2017, 13(2): e1005366.
[26]	CHEN X, HUANG Y, YOU Z H, et al. A novel approach based on KATZ measure to predict associations of human microbiota with non-infectious diseases[J]. Bioinformatics, 2017, 33(5): 733-739.
[27]	CHEN X. KATZLDA: KATZ measure for the lncRNA-disease association prediction[J]. Scientific Reports, 2015, 5: 16840.
[28]	DENG L, HUANG Y B, LIU X J, et al. Graph2MDA: a multi-modal variational graph embedding model for predicting microbe-drug associations[J]. Bioinformatics, 2022, 38(4): 1118-1125.
[29]	JIANG H J, HUANG Y, YOU Z H. Predicting drug-disease associations via using Gaussian interaction profile and kernel-based autoencoder[J]. BioMed Research International, 2019, 2019: 2426958.
[30]	YANG H P, DING Y J, TANG J J, et al. Inferring human microbe-drug associations via multiple kernel fusion on graph neural network[J]. Knowledge-Based Systems, 2022, 238: 107888.
[31]	WANG C C, LI T H, HUANG L, et al. Prediction of potential miRNA-disease associations based on stacked autoencoder[J]. Briefings in Bioinformatics, 2022, 23(2): bbac021.
[32]	LIU D Y, HUANG Y B, NIE W J, et al. SMALF: miRNA-disease associations prediction based on stacked autoencoder and XGBoost[J]. BMC Bioinformatics, 2021, 22(1): 219.
[33]	WANG S D, LIN B Y, ZHANG Y Y, et al. SGAEMDA: predicting miRNA-disease associations based on stacked graph autoencoder[J]. Cells, 2022, 11(24): 3984.
[34]	LI J, ZHANG S, LIU T, et al. Neural inductive matrix completion with graph convolutional networks for miRNA-disease association prediction[J]. Bioinformatics, 2020, 36(8): 2538-2546.
[35]	LONG Y H, LUO J W. Association mining to identify microbe drug interactions based on heterogeneous network embedding representation[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(1): 266-275.
[36]	CHENG X L, QU J, SONG S B, et al. Neighborhood-based inference and restricted Boltzmann machine for microbe and drug associations prediction[J]. PeerJ, 2022, 10: e13848.
[37]	WANG M L, CAO R Y, ZHANG L K, et al. Remdesivir and chloroquine effectively inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro[J]. Cell Research, 2020, 30(3): 269-271.
[38]	GHASEMNEJAD-BERENJI M, PASHAPOUR S. Favipiravir and COVID-19: a simplified summary[J]. Drug Research, 2021, 71(3): 166-170.
[39]	THEOHARIDES T C, CHOLEVAS C, POLYZOIDIS K, et al. Long-COVID syndrome-associated brain fog and chemofog: Luteolin to the rescue[J]. BioFactors, 2021, 47(2): 232-241.
[40]	CHOY K T, WONG A Y, KAEWPREEDEE P, et al. Remdesivir, lopinavir, emetine, and homoharringtonine inhibit SARS-CoV-2 replication in vitro[J]. Antiviral Research, 2020, 178: 104786.
[41]	DHAR J, SAMANTA J, KOCHHAR R. Corona virus disease-19 pandemic: the gastroenterologists’ perspective[J]. Indian Journal of Gastroenterology, 2020, 39(3): 220-231.