Spectral Unmixing and Compensation in Flow and Image Cytometry
Recorded On: 06/22/2019
- Non-member - Free!
- Full - Free!
- Student - Free!
- SRL Junior Staff - Free!
- SRL Emerging Leader - Free!
- Scholar - Free!
- Emeritus - Free!
- Life - Free!
- ISAC Staff - Free!
- Community Administrator - Free!
- Student Non-Member - Free!
- SRL Junior Staff Non-member - Free!
- Innovator - Free!
About the Presenters
Dr. David Novo is the president and the founder of De Novo Software Inc. De Novo Software produces FCS Express, a comprehensive, robust, and easy-to-use flow and image cytometry data analysis platform. David is a member of the ISAC Data Standards Committee and frequently consults on data analysis issues related to flow cytometry. He collaborates on many projects involving novel data analysis requirements.
Dr. Bartek Rajwa is a research associate professor of computational life sciences in the Bindley Bioscience Center at Purdue University. He conducts studies on the technology of high-throughput cytometry, high-content imaging, biological image analysis, biological pattern recognition, and applications of statistical machine learning in cell biology. Bartek is an associate editor of Cytometry Part A and a member of the Board of Directors for the Society of Biomolecular Imaging and Informatics (SBI2).
The goal of this tutorial is to present and explain the mathematical underpinnings and the physical constraints of the spectral data analysis techniques commonly employed in the fields of multidimensional flow and image cytometry. The instructors will reintroduce and thoroughly review the concepts of photon-counting uncertainty, Poissonian, sub-Poissonian, super-Poissonian statistics, spectral overlap, linear mixing models, compensation, and spectral unmixing, as understood and applied in the context of polychromatic flow cytometry, multispectral flow cytometry, and modern spectral imaging microscopy.
The tutorial will also compare and contrast the conventional compensation paradigm with the more advanced unmixing approaches, focusing on how the implicit assumptions regarding noise may produce sub-optimal results. The material will cover the impact of noise and uncertainty on collected spectral data and, consequently, on subsequent modeling, analysis, visualization, and interpretation of unmixed results. The tutorial will describe and illustrate the most commonly utilized approaches to spectral data analysis, ranging from the basic unconstrained linear unmixing with least squares minimization to blind unmixing using a family of non-negative matrix factorization heuristics. The presentation will also briefly explore the applicability of principal component analysis and other dimensionality-reduction techniques for the exploratory preview of spectral data. The intended audience includes flow and image cytometry practitioners working with all types of optical single-cell analysis platforms such as traditional polychromatic cytometers, new multispectral machines, and various spectral imaging instruments. The tutorial assumes an intermediate level of understanding of modern cytometry techniques, as well as a basic knowledge of optics and photonics.
- Basic principles of light.
- Brief introduction to fluorescence spectroscopy.
- Photon-counting statistics, photon interactions, photon collection, shot noise Poissonian, super-Poissonian, and sub-Poissonian light.
- Theory of photodetection PMTs, APDs, CCDs.
- Multispectral data-acquisition platforms in flow cytometry and imaging.
- Linear spectral mixture analysis (LSMA).
- Spectral overlap and linear mixing model.
- Use of compensation in polychromatic cytometry.
- Use of unconstrained LSMA in multispectral and hyperspectral systems.
- Abundance and non-negativity-constrained LSMA.
- Noise models in LSMA.
- Blind signal unmixing.
- Spectral data visualization and dimensionality reduction.
- Questions and open discussion.
CMLE Credit: 1.5