Introduction

Background

Phenotype-based screening of chemicals is one of the best ways to make drug discovery and chemical hazard analysis. And an off-the-shelf method is to interfere zebrafish or embryos with chemical candidates to get a quick toxicological evaluation and then conduct further deep investigations.

Zebrafish have typical advantages in quick response to chemical Dose-effects reaction, high gene similarity (over 85%) with human, rapid embryo development and so on. More importantly, most of detected unnormal phenotypes during the zebrafish development can directly relevant to the development of backboned animals. And these phenotype-driven data can have direct conversion to mammals concerning the toxic side effects of chemicals.

However, despite these advantages of zebrafish and the related literature grows, the works for phenotype-driven toxicity of chemicals concerns chemical individuals rather than systematic group analysis and most of these works cannot explain the relations between chemical structures and toxicity or phenotypes.

...

Services for users

In this work, we mainly utilize Graph Convolutional Neural Networks (GCNN) to decode chemical structures to features to predict phenotypes induced by chemicals screened by literature. And by GCNN visualizations, we can detect the chemical substructures that mostly related to phenotypes.

What the server provide?

• 5 types of toxicity prediciton on zebrafish of chemical;
• Collected chemical infomations
• Chemical Toxicity Database
• Food Additive Database

The models aim at exploring the toxicity of chemicals,and they are trained on chemicals that have been tested on zebrafish. The server provides five types of toxicity predictions(including Cardiac edema,Body Malformation,Yolk Sac Edema,Neural Toxicity and Cardiac Toxicity) on zebrafish.