Assessment of embryotoxic effects of quinoline yellow using attention-based convolutional neural network and machine learning in zebrafish model

TL;DR

This study assesses the embryotoxic effects of the synthetic dye Quinoline Yellow using zebrafish and machine learning to evaluate its potential health risks.

Contribution

The study introduces an attention-based CNN and transfer learning to automate and improve the analysis of zebrafish embryo toxicity.

Findings

QY exposure caused lethal effects and developmental abnormalities in zebrafish embryos at concentrations above 0.5 mg/mL.

In silico predictions showed QY has clastogenic, reproductive, and allergenic potential.

AI-based analysis confirmed QY's toxicity and teratogenic effects in zebrafish embryos.

Abstract

Our daily diet often includes food additives found in numerous processed foods. Growing concerns about the toxicity and potential health risks of synthetic dyes have drawn increased attention from researchers and regulatory authorities. This study examines the embryotoxic effects of Quinoline Yellow (QY), a synthetic dye commonly used as an additive, using both in silico and in vivo models. Computational studies on QY were conducted using QSAR (Quantitative Structure Activity Relations) analysis to identify the major toxicological endpoints. In silico predictions indicated clastogenic and reproductive toxicities, interaction with androgen and estrogen receptors, and an elevated propensity for skin and respiratory allergies. Danio rerio (zebrafish) embryos were exposed to various concentrations of QY (0.005–2 mg⋅mL−1) over 48, 72 and 96-h periods. Lethal effects were observed at concentrations above 0.5 mg mL−1, with a median lethal concentration LC50 of 0.64 mg mL−1. Exposure to QY (0.5–2 mg⋅mL−1) resulted in pericardial edema, swollen and necrosed yolk sac, blood stasis and reduced eye size. The study provides direct evidence for the developmental toxicity and teratogenic potential of QY. To enhance the analysis, attention-based Convolutional Neural Networks (CNN) and Transfer Learning (TL) were employed to discern morphological alterations in zebrafish embryos exposed and not exposed to QY. Automating the analysis and classification of zebrafish embryo images diminishes the workload and time burden on biological experts while simultaneously enhancing the reproducibility and objectivity of the classification. The developed neural network further corroborates the evidence suggesting QY’s potential toxicity.

Flowchart showing the procedure for assessing Quinoline Yellow using a zebrafish model. It illustrates the zebrafish developmental stages including zygote, cleavage, blastula, gastrula, segmentation, pharyngula, and hatching. Quinoline Yellow is tested and analyzed with QSAR, followed by AI-based data processing to predict toxicity, carcinogenicity, mutagenicity, and genotoxicity.

Figures