Harnessing Semi-Supervised Machine Learning to Automatically Predict Bioactivities of Per-and Polyfluoroalkyl Substances (PFASs)

Aug 3, 2026

Upload by Aurora

Per- and polyfluoroalkyl substances (PFASs) present serious health hazards because of their bioactive, persistent, and bioaccumulative behavior, yet evaluating PFAS bioactivity is slow and expensive owing to large-scale in vivo and in vitro experimentation. A semi-supervised learning framework is used to automatically predict PFAS bioactivities across multiple human biological targets, including enzymes, genes, proteins, and cell lines. The method applies semi-supervised metric learning to OECD-listed PFAS species (4730 compounds) and provides a structure–activity relationships analysis for identifying plausible bioactivities across PFAS varieties.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Abstract

Introduction

Background on PFAS hazards and exposure

Need for predictive QSAR models

Limitations of prior supervised machine-learning approaches

Proposed semi-supervised metric learning framework

UC Riverside UC Riverside Previously Published Works Title Harnessing Semi-Supervised Machine Learning to Automatically Predict Bioactivities of Perand Polyfluoroalkyl Substances (PFASs) . Permalink [https://escholarship.org/uc/item/521171r6](https://escholarship.org/uc/item/521171r6) Journal Environmental Science & Technology Letters, 10(11) ISSN 2328-8930 Authors Kwon, Hyuna Ali, Zulfikhar Wong, Bryan Publication Date 2023-11-14 DOI 10.1021/acs.estlett.2c00530 Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California This article is licensed under CC-BY 4.0 [pubs.acs.org/journal/estlcu](pubs.acs.org/journal/estlcu) Letter Harnessing Semi-Supervised Machine Learning to Automatically Predict Bioactivities of Per-and Polyfluoroalkyl Substances (PFASs) Hyuna Kwon, Zulfikhar A. Ali, and Bryan M. Wong* Cite This: Environ. Sci. Technol. Lett. 2023, 10, 1017−1022 Read Online | | | | | | | | --- | --- | --- | --- | --- | --- | | ACCESS | Metrics & More | | Article Recommendations | | *sı Supporting Information | ABSTRACT: Many per- and polyfluoroalkyl substances (PFASs) pose significant health hazards due to their bioactive and persistent bioaccumulative properties. However, assessing the bioactivities of PFASs is both timeconsuming and costly due to the sheer number and expense of in vivo and in vitro biological experiments. To this end, we harnessed new unsupervised/semisupervised machine learning models to automatically predict bioactivities of PFASs in various human biological targets, including enzymes, genes, proteins, and cell lines. Our semi-supervised metric learning models were used to predict the bioactivity of PFASs found in the recent Organisation of Economic Cooperation and Development (OECD) report list, which contains 4730 PFASs used in a broad range of industries and consumers. Our work provides the first semi-supervised machine learning study of structure−activity relationships for predicting possible bioactivities in a variety of PFAS species. KEYWORDS: per- and polyfluoroalkyl substances, PFAS, machine learning, bioactivity, semi-supervised learning ■ INTRODUCTION Since the 1930s, 1 per-and polyfluoroalkyl substances (PFASs) have been used in several consumer products (including firefighting foams) due to their outstanding stability and water/oil repellant properties.2 However, these compounds pose significant risks to the environment and biosystems. The presence of PFASs in surface water and groundwater can result in exposure to organisms, subsequently leading to accumulation in the body, with adverse effects on the liver, kidneys, blood, and immune system.2,3 Because of these deleterious effects, there is a pressing need to identify and understand the bioactivity of PFAS-based compounds that can adversely affect human health. For these reasons, several international groups including the Organisation of Economic Co-operation and Development (OECD), United States Environmental Protection Agency, Food and Drug Administration, European Chemicals Agency, European Food Safety Authority, and Ministry of Ecology and Environment (China) continue to monitor PFASs that are produced in the global market.4,5 According to a 2018 OECD report, more than 4700 PFASs currently exist as manufacturers bring new forms of PFASs into industrial and consumer products (it is worth pointing out, however, that not all 4700 structures exist in commerce). Nevertheless, among the wide varieties of PFAS molecules, the potential hazards of these new forms remain largely unknown. Due to the sheer number of PFAS species, in vivo and in vitro biological experiments are both time-consuming and costly. As such, the construction of predictive and reliable quantitative- structure activity relationship (QSAR) models6−8 is essential for assessing the bioactivities of these contaminants (even for PFAS species that are

FAQs

Why is bioactivity assessment for PFASs difficult?

PFAS bioactivity evaluation requires extensive in vivo and in vitro experiments, making it time-consuming and costly due to the large number and variety of PFAS compounds.

What machine learning approach is used to predict PFAS bioactivities?

Semi-supervised metric learning models are used to learn structure–bioactivity relationships while leveraging mostly unlabeled data with limited labeled examples.

Which PFAS dataset and targets does the study focus on?

Predictions are made for PFASs from a recent OECD report list containing 4730 compounds, with bioactivities predicted across targets such as enzymes, genes, proteins, and cell lines.

Aug 3, 2026

Upload by Aurora