全氟及多氟烷基物質（per- and polyfluoroalkyl substances, PFAS）因具有環境持久性與生物累積性，對環境及人體健康造成重大風險。然而，利用液相層析高解析質譜進行PFAS鑑定時，仍受限於參考圖譜資料庫涵蓋度不足，以及診斷碎片判讀耗時且高度仰賴人工經驗。本研究開發Neural Per- and Polyfluoroalkyl Substances Mass Spectrometry（NPFAS-MS），此為一套基於轉移學習的人工智慧模型，可預測PFAS串聯質譜圖，並建立用於環境樣本註釋的虛擬串聯質譜圖譜資料庫。研究以預訓練的CFM-ID 4.0模型為基礎，使用來自公開圖譜資料庫及實驗室分析的415張MS/MS圖譜進行微調與評估，共涵蓋140種PFAS結構，並採用MS-Clustering策略前處理以改善資料不平衡問題。最佳化後的模型針對美國環保署CompTox及NORMAN資料庫中的10,553種PFAS結構，建立包含31,659張預測圖譜的虛擬資料庫。NPFAS-MS在多項圖譜相似度指標上顯著優於現有圖譜預測模型（p < 0.001），且在資料庫檢索中達到最高的top-1召回率71.1%。應用此虛擬圖譜資料庫於水成膜泡沫滅火劑產品及地下水樣本，分別註釋了38種與40種潛在的PFAS，其中也包含新興PFAS，透過此方法降低對實驗參考質譜及人工判讀的依賴。NPFAS-MS已建置為網頁工具(https://cosbi10.ee.ncku.edu.tw/NPFAS_MS/)，支援PFAS MS/MS圖譜預測與虛擬圖譜資料庫比對，以協助大規模且快速的PFAS篩查。

Per- and polyfluoroalkyl substances (PFAS) pose substantial environmental and human health risks because of their persistence and bioaccumulation. However, PFAS identification by liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) remains limited by incomplete spectral databases and labor-intensive diagnostic fragment interpretation. This study developed Neural Per- and Polyfluoroalkyl Substances Mass Spectrometry (NPFAS-MS), a transfer learning-based artificial intelligence model for predicting PFAS tandem mass spectra and constructing a virtual spectral library for environmental annotation. A pretrained CFM-ID 4.0 model was fine-tuned and evaluated using 415 MS/MS spectra representing 140 PFAS structures from public databases and laboratory-generated data. An MS-Clustering strategy preprocessing was applied to mitigate data imbalance. The optimized model generated a virtual library containing 31,659 spectra for 10,553 PFAS structures collected from the U.S. EPA CompTox and NORMAN databases. NPFAS-MS significantly outperformed existing spectral prediction models across multiple similarity metrics (p < 0.001) and demonstrated the highest top-1 recall of 71.1% in library searching. Application of the virtual library enabled annotation of 38 and 40 PFAS candidates, including emerging PFAS, in aqueous film-forming foam (AFFF) products and groundwater samples, respectively. This approach reduced dependence on experimentally acquired reference spectra and manual interpretation. NPFAS-MS was implemented as a web-based platform (https://cosbi10.ee.ncku.edu.tw/NPFAS_MS/) for PFAS MS/MS spectrum prediction and virtual-library searching, supporting large-scale PFAS screening.

Ateia M, Maroli A, Tharayil N, Karanfil T. 2019. The overlooked short- and ultrashort-chain poly- and perfluorinated substances: A review. Chemosphere 220:866-882.
Barzen-Hanson KA, Roberts SC, Choyke S, Oetjen K, McAlees A, Riddell N, et al. 2017. Discovery of 40 classes of per- and polyfluoroalkyl substances in historical aqueous film-forming foams (AFFFs) and AFFF-impacted groundwater. Environ Sci Technol 51(4):2047-2057.
Beck AG, Muhoberac M, Randolph CE, Beveridge CH, Wijewardhane PR, Kenttämaa HI, et al. 2024. Recent developments in machine learning for mass spectrometry. ACS Meas Sci Au 4(3):233-246.
Bugsel B, Zwiener C. 2020. LC-MS screening of poly- and perfluoroalkyl substances in contaminated soil by Kendrick mass analysis. Anal Bioanal Chem 412(20):4797-4805.
Charbonnet JA, McDonough CA, Xiao F, Schwichtenberg T, Cao D, Kaserzon S, et al. 2022. Communicating confidence of per- and polyfluoroalkyl substance identification via high-resolution mass spectrometry. Environ Sci Technol Lett 9:473-481.
Chen B, Li H, Huang R, Tang Y, Li F. 2024. Deep learning prediction of electrospray ionization tandem mass spectra of chemically derived molecules. Nat Commun 15:8396.
Chen YC, Hsu JF, Chang CW, Li SW, Yang YC, Chao MR, et al. 2023. Connecting chemical exposome to human health using high-resolution mass spectrometry-based biomonitoring: Recent advances and future perspectives. Mass Spectrom Rev 42:2466-2486.
Chen YC, Wu HY, Zhuang MN, Yi CM, Wu WS, Liao PC. 2026. Enhanced environmental PFAS characterization using a virtual high-resolution mass spectral library generated by transfer learning-based neural network. Environ Sci Technol 60(24):17341-17354.
Dlugogorski BZ, Schaefer TH. 2021. Compatibility of aqueous film-forming foams (AFFF) with sea water. Fire Saf J 120:103288.
Domingo JL, Nadal M. 2019. Human exposure to per- and polyfluoroalkyl substances (PFAS) through drinking water: A review of the recent scientific literature. Environ Res 177:108648.
Evich MG, Davis MJB, McCord JP, Acrey B, Awkerman JA, Knappe DRU, et al. 2022. Per- and polyfluoroalkyl substances in the environment. Science 375:eabg9065.
Frank AM, Bandeira N, Shen Z, Tanner S, Briggs SP, Smith RD, et al. 2008. Clustering millions of tandem mass spectra. J Proteome Res 7(1):113-122.
Gebbink WA, van Asseldonk L, van Leeuwen SPJ. 2017. Presence of emerging per- and polyfluoroalkyl substances (PFASs) in river and drinking water near a fluorochemical production plant in the Netherlands. Environ Sci Technol 51(19):11057-11065.
Getzinger GJ, Higgins CP, Ferguson PL. 2021. Structure database and in silico spectral library for comprehensive suspect screening of per- and polyfluoroalkyl substances (PFASs) in environmental media by high-resolution mass spectrometry. Anal Chem 93(5):2820-2827.
Gluge J, Scheringer M, Cousins IT, DeWitt JC, Goldenman G, Herzke D, et al. 2020. An overview of the uses of per- and polyfluoroalkyl substances (PFAS). Environ Sci Process Impacts 22:2345-2373.
Horai H, Arita M, Kanaya S, Nihei Y, Ikeda T, Suwa K, et al. 2010. MassBank: A public repository for sharing mass spectral data for life sciences. J Mass Spectrom 45(7):703-714.
Huan T, Tang C, Li R, Shi Y, Lin G, Li L. 2015. MyCompoundID MS/MS Search: Metabolite identification using a library of predicted fragment-ion spectra of 383,830 possible human metabolites. Anal Chem 87(20):10619-10626.
Jiao Z, Taniyasu S, Yu N, Wang X, Yamashita N, Wei S. 2025. Two-layer homolog network approach for PFAS nontarget screening and retrospective data mining. Nat Commun 16(1):688.
Kurwadkar S, Dane J, Kanel SR, Nadagouda MN, Cawdrey RW, Ambade B, et al. 2022. Per- and polyfluoroalkyl substances in water and wastewater: A critical review of their global occurrence and distribution. Sci Total Environ 809:151003.
Lai Y, Koelmel JP, Walker DI, Price EJ, Papazian S, Manz KE, et al. 2024. High-resolution mass spectrometry for human exposomics: Expanding chemical space coverage. Environ Sci Technol 58:12784-12822.
Li Y, Kind T, Folz J, Vaniya A, Mehta SS, Fiehn O. 2021. Spectral entropy outperforms MS/MS dot product similarity for small-molecule compound identification. Nat Methods 18(12):1524-1531.
Li Y, Yao J, Zhang J, Pan Y, Dai J, Ji C, et al. 2022. First report on the bioaccumulation and trophic transfer of perfluoroalkyl ether carboxylic acids in an estuarine food web. Environ Sci Technol 56:6046-6055.
Liu Y, D'Agostino LA, Qu G, Jiang G, Martin JW. 2019. High-resolution mass spectrometry (HRMS) methods for nontarget discovery and characterization of poly- and per-fluoroalkyl substances (PFASs) in environmental and human samples. TrAC Trends Anal Chem 121:115420.
MassBank of North America. 2025. MoNA: MassBank of North America. Available: https://mona.fiehnlab.ucdavis.edu/ [accessed 18 June 2026].
Megson D, Bruce-Vanderpuije P, Idowu IG, Ekpe OD, Sandau CD. 2025. A systematic review for non-targeted analysis of per- and polyfluoroalkyl substances (PFAS). Sci Total Environ 960:178240.
NORMAN Substance Database. 2025. NORMAN SusDat. Available: https://www.norman-network.com/nds/susdat/ [accessed 18 June 2026].
Nowatzky Y, Russo FF, Lisec J, Kister A, Reinert K, Muth T, et al. 2025. Local neighborhood-based prediction of compound mass spectra from single fragmentation events. Nat Commun 16(1):2298.
Pan SJ, Yang Q. 2010. A survey on transfer learning. IEEE Trans Knowl Data Eng 22(10):1345-1359.
Patterson DE, Cramer RD, Ferguson AM, Clark RD, Weinberger LE. 1996. Neighborhood behavior: A useful concept for validation of molecular diversity descriptors. J Med Chem 39(16):3049-3059.
Ragland JM, Place BJ. 2024. A portable and reusable database infrastructure for mass spectrometry, and its associated toolkit (the DIMSpec project). J Am Soc Mass Spectrom 35(6):1282-1291.
Roth K, Yang Z, Agarwal M, Liu W, Peng Z, Long Z, et al. 2021. Exposure to a mixture of legacy, alternative, and replacement per- and polyfluoroalkyl substances (PFAS) results in sex-dependent modulation of cholesterol metabolism and liver injury. Environ Int 157:106843.
Ruttkies C, Schymanski EL, Wolf S, Hollender J, Neumann S. 2016. MetFrag relaunched: Incorporating strategies beyond in silico fragmentation. J Cheminform 8:3.
Schur C, Gasser L, Perez-Cruz F, Schirmer K, Baity-Jesi M. 2023. A benchmark dataset for machine learning in ecotoxicology. Sci Data 10(1):718.
Schymanski EL, Jeon J, Gulde R, Fenner K, Ruff M, Singer HP, et al. 2014. Identifying small molecules via high resolution mass spectrometry: Communicating confidence. Environ Sci Technol 48(4):2097-2098.
Sun J, Liu Y, Yao L, Guo Y, Ma C, Xiang T, et al. 2025. Suspect and nontarget analysis of per- and polyfluoroalkyl substances in groundwater underlying different land-use areas. Environ Sci Technol 59(5):2722-2731.
Sunderland EM, Hu XC, Dassuncao C, Tokranov AK, Wagner CC, Allen JG. 2019. A review of the pathways of human exposure to poly- and perfluoroalkyl substances (PFASs) and present understanding of health effects. J Expo Sci Environ Epidemiol 29:131-147.
Tsugawa H, Cajka T, Kind T, Ma Y, Higgins B, Ikeda K, et al. 2015. MS-DIAL: Data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nat Methods 12(6):523-526.
U.S. EPA. 2025. Method 1633, Revision A: Analysis of per- and polyfluoroalkyl substances (PFAS) in aqueous, solid, biosolids, and tissue samples by LC-MS/MS. Washington, DC: U.S. Environmental Protection Agency.
Vinaixa M, Schymanski EL, Neumann S, Navarro M, Salek RM, Yanes O. 2016. Mass spectral databases for LC/MS- and GC/MS-based metabolomics: State of the field and future prospects. TrAC Trends Anal Chem 78:23-35.
Wang F, Liigand J, Tian S, Arndt D, Greiner R, Wishart DS. 2021a. CFM-ID 4.0: More accurate ESI-MS/MS spectral prediction and compound identification. Anal Chem 93(34):11692-11700.
Wang F, Pasin D, Skinnider MA, Liigand J, Kleis JN, Brown D, et al. 2023. Deep learning-enabled MS/MS spectrum prediction facilitates automated identification of novel psychoactive substances. Anal Chem 95(50):18326-18334.
Wang H, Kuo TC, Tseng YJ. 2025. DeePFAS: Deep-learning-enabled rapid annotation of PFAS: Enhancing nontargeted screening through spectral encoding and latent space analysis. Environ Sci Technol 59(46):24841-24852.
Wang L, Yang T, Liu X, Liu J, Liu W. 2024. Critical evaluation and meta-analysis of ecotoxicological data on per- and polyfluoroalkyl substances (PFAS) in freshwater species. Environ Sci Technol 58:17555-17566.
Wang Z, Buser AM, Cousins IT, Demattio S, Drost W, Johansson O, et al. 2021b. A new OECD definition for per- and polyfluoroalkyl substances. Environ Sci Technol 55:15575-15578.
Wei JN, Belanger D, Adams RP, Sculley D. 2019. Rapid prediction of electron-ionization mass spectrometry using neural networks. ACS Cent Sci 5(4):700-708.
Williams AJ, Grulke CM, Edwards J, McEachran AD, Mansouri K, Baker NC, et al. 2017. The CompTox Chemistry Dashboard: A community data resource for environmental chemistry. J Cheminform 9(1):61.
Yan PF, Dong S, Pennell KD, Capiro NL. 2024. A review of the occurrence and microbial transformation of per- and polyfluoroalkyl substances (PFAS) in aqueous film-forming foam (AFFF)-impacted environments. Sci Total Environ 927:171883.