孤独症儿童的肠道菌群特征分析和筛查模型构建

Characteristic analysis of gut microbiota and screening model construction in children with autism spectrum disorder

摘要

目的 探讨孤独症谱系障碍(ASD)患儿和健康儿童之间肠道菌群组成的差异,并使用机器学习算法构建疾病筛查模型,提供基于肠道菌群生物标志物的非侵入性孤独症筛查手段。方法 本研究于2019年12月至2023年4月,在济南市、遵义市、香港特别行政区以及上海市招募149名2.5~4.5岁孤独症儿童为孤独症组,按年龄、性别1∶1匹配的149名健康儿童为对照组,采集粪便样本,通过16S rRNA基因V3-V4区测序收集两组儿童肠道菌群相关指标.在属水平,使用随机森林、支持向量机、K近邻算法、朴素贝叶斯分类器4种机器学习算法在模型开发数据集中构建孤独症分类模型,识别最具判别性的菌属组合,并评估模型在两个独立外部测试数据集中的泛化能力。结果 ①孤独症组的菌群多样性显著高于对照组(Chao指数=118.00、105.00,Shannon指数=3.46、3.00,P=0.023、0.001).②孤独症儿童和对照儿童肠道菌群结构存在显著差异(F=5.198,R²=0.052,P<0.001).③共筛选出14个特征菌属.其中孤独症组中丰度较高的菌属为Phocaeicola、Anaerobutyricum、Faecalibacterium、Blautia、Oscillibacter、Lachnospira、Parabacteroides、Flintibacter和Anthropogastromicrobium,对照组中丰度较高的菌属为Ruthenibacterium、Flavonifractor、Bifidobacterium、Anaerostipes和Eisenbergiella.④基于14个菌属组合的随机森林模型在模型开发数据上具有最优分类性能,训练集中曲线下面积(AUC)为100%(95%CI:100%~100%),验证集中AUC为93.94%(95%CI:88.13%~99.74%).在两个独立的外部测试集中,朴素贝叶斯模型则展现出最佳的泛化性能,AUC分别为63.83%(95%CI:51.99%~75.67%)和60.19%(95%CI:47.83%~72.55%)。结论 孤独症和健康儿童肠道微生物群落存在显著差异,且特定肠道生物标志物对孤独症疾病状态具有分类能力,提示肠道微生物具有作为儿童早期孤独症无创筛查标志物的潜在作用.

Abstract

Objective To investigate the differences of gut microbiota composition between children with autism spectrum disorder (ASD) and health children, and to construct a disease screening model using machine learning algorithm to provide a non-invasive method for autism screening based on biomarkers of gut microbiota. Methods From December 2019 to April 2023, this study recruited 149 ASD children aged 2.5 to 4.5 years from Jinan, Zunyi, Hong Kong and Shanghai, as the autism group. Additionally, 149 healthy children matched 1∶1 by age and gender were recruited as the control group. Fecal samples were collected, and gut microbiota-related indices were gathered through 16S rRNA gene V3-V4 region sequencing for both groups. At the genus level, four machine learning algorithms, random forest, support vector machine, K-nearest neighbors, and naive bayes classifier, were used to construct an autism classification model in the model development dataset, identifying the most discriminative bacterial genus combinations, and the generalization ability of the models was evaluated in two independent external test datasets. Results ①The gut microbiota diversity of the autism group was significantly higher than that in the control group (Chao index=118.00, 105.00; Shannon index=3.46, 3.00; P=0.023, 0.001). ②There were significant differences in gut microbiota structure between autism children and control children (F=5.198, R²=0.052, P<0.001). ③A total of 14 characteristic genera were identified. The genera with higher abundance in the autism group were Phocaeicola, Anaerobutyricum, Faecalibacterium, Blautia, Oscillibacter, Lachnospira, Parabacteroides, Flintibacter, and Anthropogastromicrobium, and the genera with higher abundance in the control group were Ruthenibacterium, Flavonifractor, Bifidobacterium, Anaerostipes, and Eisenbergiella. ④The random forest model based on the combination of 14 genera showed the best classification performance in the model development dataset, with the training set AUC of 100% (95%CI: 100%-100%) and validation set AUC of 93.94% (95%CI: 88.13%-99.74%). In two independent external test datasets, the Naive Bayes model showed the best generalization performance, with AUC of 63.83% (95%CI: 51.99%-75.67%) and 60.19% (95%CI: 47.83%-72.55%), respectively. Conclusion There are significant differences in the gut microbiota communities between autism children and control children, and specific gut microbiota biomarkers have the capability to classify autism disease states, suggesting that gut microbiota has potential significance as a non-invasive screening biomarker for early autism detection in children.

关键词

肠道菌群 / 孤独症 / 儿童 / 生物标志物 / 机器学习 / 16S rRNA

Key words

gut microbiota / autism / children / biomarkers / machine learning / 16S rRNA

参考文献

[1] Rosen N E, Lord C, Volkmar F R. The diagnosis of autism: from Kanner to DSM-III to DSM-5 and beyond[J]. J Autism Dev Disord, 2021, 51(12): 4253-4270.

[2] Zeidan J, Fombonne E, Scorah J, et al. Global prevalence of autism: a systematic review update[J]. Autism Res, 2022, 15(5): 778-790.

[3] Maenner M J, Warren Z, Williams A R, et al. Prevalence and characteristics of autism spectrum disorder among children aged 8 years - autism and developmental disabilities monitoring network, 11 sites, United States, 2020[J]. MMWR Surveill Summ, 2023, 72(2): 1-14.

[4] 贾美香. 提升对我国孤独症谱系障碍儿童发病状况及早期干预的关注[J]. 中国妇幼健康研究, 2023, 34(1): 1-4.

[5] Wan Y, Zuo T, Xu Z, et al. Underdevelopment of the gut microbiota and bacteria species as non-invasive markers of prediction in children with autism spectrum disorder[J]. Gut, 2022, 71(5): 910-918.

[6] Bundgaard-Nielsen C, Knudsen J, Leutscher P D C, et al. Gut microbiota profiles of autism spectrum disorder and attention deficit/hyperactivity disorder: a systematic literature review[J]. Gut Microbes, 2020, 11(5): 1172-1187.

[7] Cho H, Qu Y, Liu C, et al. Comprehensive evaluation of methods for differential expression analysis of metatranscriptomics data[J]. Brief Bioinform, 2023, 24(5): bbad279.

[8] Jayasinghe W J M L P, Deo R C, Ghahramani A, et al. Deep multi-stage reference evapotranspiration forecasting model: multivariate empirical mode decomposition integrated with the boruta-random forest algorithm[J]. Ieee Access, 2021, 9: 166695-166708.

[9] Monteith S, Glenn T, Geddes J, et al. Expectations for artificial intelligence (AI) in psychiatry[J]. Curr Psychiatry Rep, 2022, 24(11): 709-721.

[10] Wang M, Doenyas C, Wan J, et al. Virulence factor-related gut microbiota genes and immunoglobulin a levels as novel markers for machine learning-based classification of autism spectrum disorder[J]. Comput Struct Biotechnol J, 2021, 19: 545-554.

[11] Gomez-Nguyen A, Basson A R, Dark-Fleury L, et al. Parabacteroides distasonis induces depressive-like behavior in a mouse model of Crohn's diseas[J]. Brain Behav Immun, 2021, 98: 245-250.

[12] Inoue R, Sakaue Y, Sawai C, et al. A preliminary investigation on the relationship between gut microbiota and gene expressions in peripheral mononuclear cells of infants with autism spectrum disorders[J]. Biosci Biotechnol Biochem, 2016, 80(12): 2450-2458.

[13] Li R, Li L, Hong P, et al. β-Carotene prevents weaning-induced intestinal inflammation by modulating gut microbiota in piglets[J]. Anim Biosci, 2021, 34(7): 1221-1234.

[14] Kircher B, Woltemate S, Gutzki F, et al. Predicting butyrate-and propionate-forming bacteria of gut microbiota from sequencing data[J]. Gut Microbes, 2022, 14(1): 2149019.

[15] Ritz N L, Brocka M, Butler M I, et al. Social anxiety disorder-associated gut microbiota increases social fear[J]. Proc Natl Acad Sci USA, 2024, 121(1): e2308706120.

[16] Zhao Y, Wang Y, Meng F, et al. Altered gut microbiota as potential biomarkers for autism spectrum disorder in early childhood[J]. Neuroscience, 2023, 523: 118-131.

[17] Morton J T, Jin D M, Mills R H, et al. Multi-level analysis of the gut-brain axis shows autism spectrum disorder-associated molecular and microbial profiles[J]. Nat Neurosci, 2023, 26(7): 1208-1217.

[18] Charilaou P, Battat R. Machine learning models and over-fitting considerations[J]. World J Gastroenterol, 2022, 28(5): 605-607.

[19] Collins G S, Reitsma J B, Altman D G, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement[J]. BMJ, 2015, 350: g7594.

[20] Kumar Y, Koul A, Singla R, et al. Artificial intelligence in disease diagnosis: a systematic literature review, synthesizing framework and future research agenda[J]. J Ambient Intell Humaniz Comput, 2023, 14(7): 8459-8486.

[21] Chatterjee A, Pahari N, Prinz A, et al. Machine learning and ontology in eCoaching for personalized activity level monitoring and recommendation generation[J]. Sci Rep, 2022, 12(1): 19825.

基金资助

国家自然科学基金(82181220077)

国家自然科学基金(8237120466)

复旦大学公共卫生学院嘉定区卫生健康委公共卫生高质量发展重点学科、重点专项(GWGZLXK-G2023-G04)

上海市加强公共卫生体系建设三年行动计划(2023-2025年)重点学科(GWVI-G11.1-G32)