Using Machine Learning to Construct the Blood–Follicle Distribution Models of Various Trace Elements and Explore the Transport-Related Pathways with Multiomics Data

• Published in: Environmental science & technology 2024-05, Vol.58 (18), p.7743-7757
• Main Authors: Zhang, Guohuan, Lin, Weinan, Gao, Ning, Lan, Changxin, Ren, Mengyuan, Yan, Lailai, Pan, Bo, Xu, Jia, Han, Bin, Hu, Ligang, Chen, Yuanchen, Wu, Tianxiang, Zhuang, Lili, Lu, Qun, Wang, Bin, Fang, Mingliang
• Format: Article
• Language: English
• Subjects: Antimony; Arsenic; Biological analysis; Bismuth; Blood; Cadmium; China; Ecotoxicology and Public Health; Embryo transfer; Female; Follicles; Follicular fluid; Follicular Fluid - chemistry; Follicular Fluid - metabolism; Gametocytes; Germanium; Glucose; Glucose metabolism; Hormones; Humans; In vitro fertilization; Learning algorithms; Machine Learning; Manganese; Mercury; Metabolism; Multiomics; Nickel; Permeability; Proteomes; Selenium; Serum levels; Steroid hormones; Steroids; Trace elements; Trace Elements - metabolism
• ISSN: 0013-936X; 1520-5851; 1520-5851
• DOI: 10.1021/acs.est.3c10904

Permeabilities of various trace elements (TEs) through the blood–follicle barrier (BFB) play an important role in oocyte development. However, it has not been comprehensively described as well as its involved biological pathways. Our study aimed to construct a blood–follicle distribution model of the concerned TEs and explore their related biological pathways. We finally included a total of 168 women from a cohort of in vitro fertilization-embryo transfer conducted in two reproductive centers in Beijing City and Shandong Province, China. The concentrations of 35 TEs in both serum and follicular fluid (FF) samples from the 168 women were measured, as well as the multiomics features of the metabolome, lipidome, and proteome in both plasma and FF samples. Multiomics features associated with the transfer efficiencies of TEs through the BFB were selected by using an elastic net model and further utilized for pathway analysis. Various machine learning (ML) models were built to predict the concentrations of TEs in FF. Overall, there are 21 TEs that exhibited three types of consistent BFB distribution characteristics between Beijing and Shandong centers. Among them, the concentrations of arsenic, manganese, nickel, tin, and bismuth in FF were higher than those in the serum with transfer efficiencies of 1.19–4.38, while a reverse trend was observed for the 15 TEs with transfer efficiencies of 0.076–0.905, e.g., mercury, germanium, selenium, antimony, and titanium. Lastly, cadmium was evenly distributed in the two compartments with transfer efficiencies of 0.998–1.056. Multiomics analysis showed that the enrichment of TEs was associated with the synthesis and action of steroid hormones and the glucose metabolism. Random forest model can provide the most accurate predictions of the concentrations of TEs in FF among the concerned ML models. In conclusion, the selective permeability through the BFB for various TEs may be significantly regulated by the steroid hormones and the glucose metabolism. Also, the concentrations of some TEs in FF can be well predicted by their serum levels with a random forest model.