Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
Date: Tuesday 31st March 2020
Time: 14:00 - 15:00 UK Time
Speakers: Ketil Tvermosegaard (GSK).
The slides for this event can be downloaded here.
This webinar is free for PSI Members and Non-Members.
Label-based flow cytometry allows the quantification of target features of interest by attaching fluorophores (labels) to antibodies and measuring the resulting fluorescence at the relevant wavelengths. This is widely used for cell sorting, i.e., determining cell types.
Image flow cytometry is a technology which enables single cell images in cell sorting experiments. Problematically, directly using this data for classification involves manual inspection of many thousands of images. This creates a bottleneck for analysis and scalability.
As part of an epithelial barrier project; a Medium Throughput screen was conducted to investigate whether candidate CRISPR gene knockouts modulated the proportion of cells which differentiated into ciliated cells (important for indications such as COPD and asthma).
However, the team hypothesized that traditional label-based flow cytometry did not always properly classify cell types. We were approached about developing a scalable way of using image flow cytometry for determining whether cells are ciliated. This would provide them with an alternative endpoint and a way to test their hypothesis.
In this project, we;
1) Developed Python code to extract images from the proprietary file format
2) Built a proof-of-concept convolutional neural network. Results here suggested the problem was solvable with Deep Learning
3) Initiated a Tessella Analytics Partnership project with Tessella
4) Worked with Tessella to steer their development of an appropriate architecture for the neural network, which achieved better-than-human performance
5) Applied the trained network to a validation screen and confirmed disagreements between label-based and label-free flow cytometry.
