Analysis of a genome-wide microscopy screen for regulators of lysosomal stability
(2020) BINP52 20191Degree Projects in Bioinformatics
- Abstract
- Lysosomes are cell organelles in charge of degradation. Lysosomal membrane permeabilization (LMP) causes the leakage of cathepsins and other digestive enzymes into the cytosol from lysosomal cavities. This usually causes the initiation of cell death. LMP can be triggered by many stimuli and this could potentially be used in new therapeutic approaches for removing cancer cells. However, excessive LMP also contributes to numerous diseases, e.g. neurodegeneration, stroke, and some infectious diseases. There are therefore great potential therapeutic benefits in LMP regulation. However, there is currently not enough knowledge in this field because of the long lack of good assays for measuring LMP. However, a novel assay has recently been... (More)
- Lysosomes are cell organelles in charge of degradation. Lysosomal membrane permeabilization (LMP) causes the leakage of cathepsins and other digestive enzymes into the cytosol from lysosomal cavities. This usually causes the initiation of cell death. LMP can be triggered by many stimuli and this could potentially be used in new therapeutic approaches for removing cancer cells. However, excessive LMP also contributes to numerous diseases, e.g. neurodegeneration, stroke, and some infectious diseases. There are therefore great potential therapeutic benefits in LMP regulation. However, there is currently not enough knowledge in this field because of the long lack of good assays for measuring LMP. However, a novel assay has recently been developed that enabled detection of individual permeabilized lysosomes (the galectin LMP assay). My host group has now used the assay together with other cell death markers in a genome-wide high-content imaging screen to elucidate which proteins regulate LMP and cell death. The aim of this project is to analyse the screen image data set for extracting information regarding genes functioning in the cell death process. This was done by machine learning and compared to the traditional approach to analysing high-content imaging screens using CellProfiler. In the first step of the project, I explored the characteristics of the data set. Most of the images had black, non-informative background that made the detection of small, sub-cellular features, a big challenge even for the human vision. Besides, analyzing such a huge data set with that amount of redundancy in order to extract sparse information by machine learning algorithms, was practically challenging. Thus, I extracted a uniform, random subset of images and made cutouts centered on one cell to test different objectlevel analysis approaches. Unsupervised clustering with different feature reduction techniques was examined to find different types of cell death. However, no distinct clusters were observed. I therefore created an online game for citizen-science based image annotation to obtain a data set with labels for supervised learning algorithms. This is ready to be launched. When comparing new machine learning based approaches with conventional methods such as CellProfiler, we found that the new approaches were more time-consuming and challenging due to the technical issues such as the need for a large annotated training data set and super computational resources and a lack of established workflows. During the continuation of the project it will be determined if machine learning methods are nevertheless superior due to their high capability of exploring deeper and more accurate features of the data. (Less)
- Popular Abstract
- Analysing a large microscopy dataset related to cell death using artificial intelligence
Cell death is an essential process for quality control in our organism as it eliminates infected or damaged cells. Too much or too little cell death, however, is problematic and contributes to many diseases, e.g. stroke, degenerative diseases or cancer. In this project, we are trying to determine in how many ways cells can die and how the different types of cell death are regulated by analyzing several million cell images that my host group had previously taken in a large microscopy experiment. We were particularly interested in the status of the lysosomes, small compartments enclosed by a membrane, which are involved in different types of cell... (More) - Analysing a large microscopy dataset related to cell death using artificial intelligence
Cell death is an essential process for quality control in our organism as it eliminates infected or damaged cells. Too much or too little cell death, however, is problematic and contributes to many diseases, e.g. stroke, degenerative diseases or cancer. In this project, we are trying to determine in how many ways cells can die and how the different types of cell death are regulated by analyzing several million cell images that my host group had previously taken in a large microscopy experiment. We were particularly interested in the status of the lysosomes, small compartments enclosed by a membrane, which are involved in different types of cell death. Lysosomes can lose their membrane integrity and release harmful substances into the main part of the cell causing its death, which is called lysosome membrane permeabilization (LMP).
To evaluate the cell images I used different computational strategies and tools including machine learning, a type of artificial intelligence. The total size of the image set was so large (about 23 terabytes) that I could not do anything without the help of high-capacity Swedish storage and computing servers and also had to work with only a fraction of the images at this point. I first explored the characteristics of some representative images using mathematical techniques. Then, I resorted to a variety of machine learning techniques. I started with different unsupervised methods, which are methods that try to find patterns in the data. I tried to use these methods to group the cell images by similarity in order to understand which different types of dead or dying cells exist. However, no clear groups were seen with any of the techniques I used. The reason this approach was not successful could be that large parts of the images were black background.
I therefore switched to supervised methods which rely on example images classified by a “supervisor”. To obtain enough labeled images, which is too time consuming for a few biology experts, I designed a citizen science based online game, called Cell Hunter, that I will launch shortly. In the game, ordinary people will label cell images according to several cell death-related characteristics defined by biology experts. The website for this game will be shared in public so that everyone who wants can join. In the future, I will use the labelled images from the game to train artificial intelligence tools called deep neural networks that can learn to detect the different cell death characteristics and classify the entire set of several million images.
This project is a part of a larger project that aims to find gene networks that are effective in regulating cell death. This can provide new approaches to treat the many diseases that are associated with higher or lower than usual cell death rate.
Master’s Degree Project in Bioinformatics 60 credits 2019
Department of Biology, Lund University
Advisor: Sonja Aits, Unit/Department: Cell Death and Lysosomes, Faculty of Medicine/Department of Experimental Medical Science, Lund University (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9006926
- author
- Kazemi Rashed, Salma
- supervisor
-
- Sonja Aits LU
- organization
- course
- BINP52 20191
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- English
- id
- 9006926
- date added to LUP
- 2020-03-20 14:35:26
- date last changed
- 2020-03-20 14:35:26
@misc{9006926,
abstract = {{Lysosomes are cell organelles in charge of degradation. Lysosomal membrane permeabilization (LMP) causes the leakage of cathepsins and other digestive enzymes into the cytosol from lysosomal cavities. This usually causes the initiation of cell death. LMP can be triggered by many stimuli and this could potentially be used in new therapeutic approaches for removing cancer cells. However, excessive LMP also contributes to numerous diseases, e.g. neurodegeneration, stroke, and some infectious diseases. There are therefore great potential therapeutic benefits in LMP regulation. However, there is currently not enough knowledge in this field because of the long lack of good assays for measuring LMP. However, a novel assay has recently been developed that enabled detection of individual permeabilized lysosomes (the galectin LMP assay). My host group has now used the assay together with other cell death markers in a genome-wide high-content imaging screen to elucidate which proteins regulate LMP and cell death. The aim of this project is to analyse the screen image data set for extracting information regarding genes functioning in the cell death process. This was done by machine learning and compared to the traditional approach to analysing high-content imaging screens using CellProfiler. In the first step of the project, I explored the characteristics of the data set. Most of the images had black, non-informative background that made the detection of small, sub-cellular features, a big challenge even for the human vision. Besides, analyzing such a huge data set with that amount of redundancy in order to extract sparse information by machine learning algorithms, was practically challenging. Thus, I extracted a uniform, random subset of images and made cutouts centered on one cell to test different objectlevel analysis approaches. Unsupervised clustering with different feature reduction techniques was examined to find different types of cell death. However, no distinct clusters were observed. I therefore created an online game for citizen-science based image annotation to obtain a data set with labels for supervised learning algorithms. This is ready to be launched. When comparing new machine learning based approaches with conventional methods such as CellProfiler, we found that the new approaches were more time-consuming and challenging due to the technical issues such as the need for a large annotated training data set and super computational resources and a lack of established workflows. During the continuation of the project it will be determined if machine learning methods are nevertheless superior due to their high capability of exploring deeper and more accurate features of the data.}},
author = {{Kazemi Rashed, Salma}},
language = {{eng}},
note = {{Student Paper}},
title = {{Analysis of a genome-wide microscopy screen for regulators of lysosomal stability}},
year = {{2020}},
}