flowjo webinars: machine learning in flowjo v10...•intro to machine learning •supervised...
TRANSCRIPT
![Page 1: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/1.jpg)
Christian R. Aguilera-Sandoval, PhDField Applications Scientist, [email protected]
Jack Panopoulos, PhDField Applications Scientist, [email protected]
FlowJo Webinars: Machine Learning in FlowJo v10
![Page 2: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/2.jpg)
Outline
• Gating practices in flow cytometry• Manual Hierarchical gating• High-dimensionality reduction techniques and clustering algorithms
• Intro to Machine Learning• Supervised• Unsupervised
• Description of Dimredux techniques and clustering algorithms• Comparison of dimredux techniques• Comparison of clustering algorithms• Which tool should I use and when?• Live Demo of dimreduc techniques and clustering algorithms• Coming Soon: Preview of unreleased plug-in
• Cluster quality check
![Page 3: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/3.jpg)
Current Gating Practices in Flow Cytometry
• Manual hierarchical gating • High-dimensionality reduction and clustering
![Page 4: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/4.jpg)
Manual Hierarchical Gating
• Manual hierarchical gating• Traditional gating using parent and child gates• Visual inspection of bivariate plots to identify cell populations• Problematic for high-dimensional settings
• Subjectivity • Operator bias• Difficulty of identification for rare populations• Reproducibility• Populations could be overseen/ignored due to high parameter use
![Page 5: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/5.jpg)
High-dimensionality Reduction and Clustering
• Pros• Efficient and time saving for high-dimensionality panels• Uncover multidimensional structures not seen in bivariate projections• Uses machine learning as the tool to identify cell phenotypes reducing bias,
user error yet increasing reproducibility• Cons:• Technical expertise• Advanced computer hardware
• Machine learning methods• Supervised• Unsupervised
![Page 6: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/6.jpg)
Machine Learning - Supervised
![Page 7: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/7.jpg)
Machine Learning - Unsupervised
![Page 8: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/8.jpg)
FlowJo’s Machine Learning’s ToolsTo Be Discussed Today
Method Purpose
tSNE1 DimRex Method to visualize general structure of nonlinear data and assess a sample’s heterogeneity through high resolution of local neighborhood structure
UMAP2 DimRex Method to quickly visualize the local and global structure of nonlinear data as a trade-off to resolution of local neighborhood structure and fidelity of global structure
TriMAP3 DimRex Method to visualize with high fidelity the global structure of nonlinear data
FlowSOM4 Clustering Self-organizing maps, followed by hierarchical consensus meta-clustering to merge clusters
Xshift5 Clustering Weighted k-nearest neighbor density estimation, detection of density centroids, cells linked by centroid via density-ascending paths
Phenograph6 Clustering Detection of k-nearest of neighbors of each cell, followed by partitioning of the map into clusters based on phenotype
1= Maaten et al, 2008 2= McInnes et al 2018 3= Amid et al, 2019 4= Van Gassen, 2015 5=Samusik et al, 2016 6= Levine et al, 2015
![Page 9: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/9.jpg)
A Comparison of DimRex Algorithms
Amid et al, 2019
![Page 10: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/10.jpg)
Study Flowchart
Lie et al 2019
![Page 11: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/11.jpg)
LDA Is Superior to ACDC In Precision*
Liu et al 2019
![Page 12: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/12.jpg)
Unsupervised Methods Outperform Semisupervised Methods In Coherence
Lie et al 2019
![Page 13: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/13.jpg)
Best Unsupervised Clustering Methods for Inner Structure
• FlowSOM• Phenograph• DEPECHE
![Page 14: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/14.jpg)
Phenograph and LDA Are The Most RobustMethods With Repetitive Tests or Varying Sample Sizes
Lie et al 2019
![Page 15: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/15.jpg)
Phenograph Detects More Clusters Consistently In Varying Sample Sizes
![Page 16: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/16.jpg)
Xshift And Phenograph Best Idenfity Subsets Of Major Cell Types
Clus
ters
by
Phen
ogra
ph
![Page 17: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/17.jpg)
A B
C
D
Liu et al 2019; Weber et al, 2016
![Page 18: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/18.jpg)
The Right Tool For The Right Job
Lie et al 2019
![Page 19: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/19.jpg)
Conclusions
• Top unsupervised tools: • FlowSOM and Phenograph
• Precision• Coherence• Stability
• Top tools to detect subsets:• Phenograph and Xshift (using Elbow Plot Determination)
• FlowSOM best tool when dealing with very large datasets
![Page 20: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/20.jpg)
Next: Live Demo
![Page 21: FlowJo Webinars: Machine Learning in FlowJo v10...•Intro to Machine Learning •Supervised •Unsupervised •Description of Dimreduxtechniques and clustering algorithms •Comparison](https://reader035.vdocuments.us/reader035/viewer/2022062506/5f0e55597e708231d43ebc63/html5/thumbnails/21.jpg)
Thank You!!!
Christian R. Aguilera-Sandoval, [email protected]
Serena Di Cecilia, [email protected]
Jack Panopoulos, [email protected]
Tech [email protected]