Automated phenotype recognition for zebrafish embryo based in vivo high throughput toxicity screening of engineered nano-materials.

Abstract

A phenotype recognition model was developed for high throughput screening (HTS) of engineered Nano-Materials (eNMs) toxicity using zebrafish embryo developmental response classified, from automatically captured images and without manual manipulation of zebrafish positioning, by three basic phenotypes (i.e., hatched, unhatched, and dead). The recognition model was built with a set of vectorial descriptors providing image color and texture information. The best performing model was attained with three image descriptors (color histogram, representative color, and color layout) identified as most suitable from an initial pool of six descriptors. This model had an average recognition accuracy of 97.40±0.95% in a 10-fold cross-validation and 93.75% in a stress test of low quality zebrafish images. The present work has shown that a phenotyping model can be developed with accurate recognition ability suitable for zebrafish-based HTS assays. Although the present methodology was successfully demonstrated for only three basic zebrafish embryonic phenotypes, it can be readily adapted to incorporate more subtle phenotypes.