Automated and quantitative phenotyping of C. elegans genetic screens from high-throughput image data

High throughput screens producing image data, are becoming increasingly easy to perform. Nevertheless, manual evaluation of image data is impractical because it is not only time intensive, but also prone to error. The model organism C. elegans is frequently used to study fundamental questions in development and behavior and is particularly amenable to high throughput screening due to its small size and ability to be cultured in a well of a 96 well plate. In this thesis I will describe an automated image analysis system (DevStaR) for quantitative phenotyping of embryonic lethality and sterility from populations of C. elegans in 96 well plates. This image analysis system counts each developmental stage in an image of a C. elegans population, allowing efficient high throughput calculation of C. elegans viability phenotypes.;DevStaR is an object recognition machine comprising several hierarchical layers that build successively more sophisticated representations of the objects (developmental stages) to be classified. The algorithm segments the objects, decomposes the objects into parts, extracts features from these parts, and classifies them using a Support Vector Machine (SVM) and global shape information. This enables correct classifications in the presence of complicated occlusions and deformations of the animals. Features of the classified objects are then used to obtain a count of each developmental stage.;I have used this system to analyze phenotypic data from approximately 50 C. elegans genome wide genetic interaction screens, as well as a genome wide RNAi screen in high replication (~30 replicates per RNAi clone). Using the quantitative phenotype output by DevStaR I have examined features of the high-throughput screen data, such as variability and penetrance, which have not been examined in detail previously due to a lack of automated and quantitative scoring methods available for high-throughput image data.;DevStaR overcomes a previous bottleneck in image analysis by achieving near real-time scoring of image data in a fully automated manner. Moreover, DevStaR reduces the need for human evaluation of images and provides rapid quantitative output that is not feasible at high throughput by manual scoring.