-
Notifications
You must be signed in to change notification settings - Fork 2
Home
Welcome to the neural-profiling wiki! This wiki will give you an introduction to the repository and the project as a whole. For any questions, don't hesitate to contact @michaelbornholdt ([email protected]).
This project resulted in a Master thesis. I advise to look at the Methods and Conclusion section for details and an overview of the experiments presented in this repository.
This project was completed during a research internship in the Carpenter-Singh lab during the summer of 2021. My supervisors at the Broad institute are Juan Caicedo and Shantanu Singh and my supervisor from Germany is Prof. Daniel Rückert. The following people were also involved in my project and helped me at various points. Thank you to Gregory Way, Nikita Moshkov and Niranj Chandrasekaran
“I will design a protocol for robustly and scalably quantifying cellular states from images using deep neural networks; the JUMP consortium will use this protocol to profile >2B cells to generate a dataset that will be made public in 2022”
The JUMP-Cell Painting Consortium is creating a new data-driven approach to drug discovery based on cellular imaging, image analysis, and high dimensional data analytics. This project contributes to the consortium's goal by developing and exploring the pipeline of deep neural-net-driven feature space extraction.
The Library of Integrated Network-Based Cellular Signatures, short the LINCS database aims to create publicly available resources to characterize how cells respond to perturbation. Around 100 million A549 (lung cancer) cells are perturbed with 1,571 different compounds across 6 doses in 5 technical replicates. The classical profiling steps employed in the LINCS project shares many similarities with my project.
The aim of this first chapter is to decide on useful evaluation metrics and thus create a baseline of how well the class pipeline performs. Classical feature extraction is done by employing CellProfiler and following the postprocessing steps found in the LINCS repository.
In the second chapter, I use DeepProfiler, the cell images, and the masked cell locations to infer the features of each cell by running the images through a pre-trained net on DeepProfiler. After using similar postprocessing as for the CellProfiler (CP), I can compare the CP baseline to the pre-trained nets via the evaluation metrics.
Finally, I train my own models to be able to classify single-cell images into their respective perturbations. This so-called Weakly Supervised Representation Learning leads the pre-trained model to become more sensitive to cell inputs and partially displays better metric scores.