SETApp: A machine learning and image analysis based application to automate the sea urchin embryo test

• Published in: Ecotoxicology and environmental safety Vol. 241; p. 113728
• Main Authors: Alvarez-Mora, Iker, Mijangos, Leire, Lopez-Herguedas, Naroa, Amigo, Jose M., Eguiraun, Harkaitz, Salvoch, Maddi, Monperrus, Mathilde, Etxebarria, Nestor
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier Inc 01.08.2022; Elsevier
• Subjects: Ecotoxicology; Effect-directed analysis; High-throughput screening; Image analysis; Machine learning; Sea urchin embryo test; High-throughput screening; Image analysis; Sea urchin embryo test; Ecotoxicology; Effect-directed analysis; Machine learning
• ISSN: 0147-6513; 1090-2414
• DOI: 10.1016/j.ecoenv.2022.113728

Since countless xenobiotic compounds are being found in the environment, ecotoxicology faces an astounding challenge in identifying toxicants. The combination of high-throughput in vivo/in vitro bioassays with high-resolution chemical analysis is an effective way to elucidate the cause-effect relationship. However, these combined strategies imply an enormous workload that can hinder their implementation in routine analysis. The purpose of this study was to develop a new high throughput screening method that could be used as a predictive expert system that automatically quantifies the size increase and malformation of the larvae and, thus, eases the application of the sea urchin embryo test in complex toxicant identification pipelines such as effect-directed analysis. For this task, a training set of 242 images was used to calibrate the size-increase and malformation level of the larvae. Two classification models based on partial least squares discriminant analysis (PLS-DA) were built and compared. Moreover, Hierarchical PLS-DA shows a high proficiency in classifying the larvae, achieving a prediction accuracy of 84 % in validation. The scripts built along the work were compiled in a user-friendly standalone app (SETApp) freely accessible at https://github.com/UPV-EHU-IBeA/SETApp. The SETApp was tested in a real case scenario to fulfill the tedious requirements of a WWTP effect-directed analysis. [Display omitted] •A novel predictive system (SETApp) has been developed to automatically quantify the two endpoints of the sea urchin embryo test.•The usefulness of PLS-DA and image parametrization in the automation of the bioassays has been proven.•The SETApp was implemented in a demanding scenario: the effect-directed analysis of Bayonne’s (France) WWTP effluent.•The EDA study concluded that the SETApp is an efficient, fast, cost-effective and reproducible tool that can approach EDA to routine analysis.•The suspect screening identified six contaminants of emerging concern as toxicity contribution candidates.