Clinical Flow Labs in Vancouver

T-ALL

B-ALL

Immune deficiencies

ACL/AML

MM

PNH

LPD

Test menus

Clinical flow labs in Vancouver

LPD

8 Colors 3 Tubes

ACL

8 Colors 5 Tubes

LPD

13 Colors 3 Tubes

LPD

14 Colors 3 Tubes

ACL

12 Colors 5 Tubes

AUTOFLOW

LPD

30 Colors 2 Tubes

AUTOFLOW2.0

Timeline

AML MRD

12 Colors 3 Tubes

Data analysis in Flow cytometry

Flow cytometry data is high dimensional

Analysis in Low D

Automated Gating

CD8

CD4

CD4

Flow cytometry data analysis

Dimensional Reduction

Clustering

Classification

CD4 T

CD8 T

Monocytes

Analysis in High D

Malek M. et al., Bioinformatics. 31, 606–607 (2015)

Rahim A. et al., Methods. 134-135 (2018)

Automated Gating

A supervised approach to determining the optimal 'gate' locations for individual markers based on their density distributions

Perform automated gating with pre-defined gating strategy

k-Mean

flowPeaks

FlowSOM

PhenoGraph

X-shift

ACCENSE

... ...

Common Clustering Algorithms:

Clustering

Grouping similar objects

Wang X et al., Nat. Commun. 13,6772 (2022)

Categorize or distinguish objects (cells) into predefined classes

Logistic Regression

Naive Bayes

Random Forest

K-Nearest Neighbours

XG-Boost

Neural Networks

T? B? NK? Mono?

Population/ cell level:

Common Classifier:

Classification

Specimen level:

Wang X et al., Nat. Commun. 13,6772 (2022)

Dimensional Reduction 

UMAP

L. McInnes, J. Healy, J. Melville, Arxiv (2018). 

https://pair-code.github.io/understanding-umap/

PCA:

Find best angles to look at data from lower dimensions

PCA: Principle Component analysis

UMAP: Uniform Manifold Approximation and Projection

Dimensional Reduction 

UMAP: Reconstruct data in lower dimensions

Manual

Manual analysis vs Auto analysis 

Data Pre-processing

Gating

Annotation

Reporting

Data Pre-processing

Automated Gating

Classification

Data visualization

FCS modification

Clustering

Dimensional Reduction

Auto

• FlowCore

• FlowDensity

• FloWorkspace

• Scanppy/FlowSom

• FlexDashboard

• RShiny

.......

Import/export/manipulation of flow cytometry data

Perform sequential "gating"

Handling cell population and "gating" objects

Dimension reduction and clustering

Graphical user interface and user experiences

Open source tools

Automating flow data analysis to improve tests efficiencies 

From sample to report

2000

4000

Complexity

LPD

MM

CD34

PNH

ACL

Increasing test volume

high test complexities

Test volumes vs. complexities

AUTOFLOW

MAGIC-DR

• Deliver objective analysis while enhancing reproducibility

• Offer clear, intuitive, and informative presentations of flow result

• Reduce technologist and pathologist workload

We don't want a tool  that make disease diagnosis in a black box

• Human-in-the-loop and Easily Integratabtle ML framework for AML MRD analysis into conventional analysis

• Improve the efficiency and accuracy of the AML MRD analysis 

MAGIC: MRD Analysis Guided by Integrated Classifier

DR: Dimensionality  Reduction

Strategies

Data Pre-processing

Automated Gating

Classification

FCS modification

Clustering

Dimensional Reduction

GUI

Data Pre-processing

Dimensional Reduction

AUTOFLOW for

Lymphoproliferative Disorders flow

• TubeB1: B Lymphoma Clonality and  subtypes

• TubeB2: B Lymphoma subtypes

• TubeT: T Lymphoma Clonality and subtypes

LPD panel

Total

(SSC/CD45)

Mono+Lymph

(CD14/CD7)

Lymphocytes

(CD2/CD3)

T cells

(CD4/CD8)

CD4

CD8

(CD5/CD7)

CD4

(CD5/CD7)

CD4

(CD7/TRBC1)

CD8

(CD5/TRBC1)

CD8

... ...

... ...

... ...

Supervised automated gating for Tube T

Gating hierarchy

>300 populations

Supervised automated gating for Tube T

Tumoral heterogeneity makes it difficult to design a universal strategy to gate the malignant B cell populations

Analyzing TubeB1

Total

(SSC/CD45)

Mono+Lymph

(CD14/IgM)

Lymphocytes

(CD19+ or CD20+)

B cells

Unsupervised clustering (SOM)

Dimension reduction (UMAP)

Hybrid approach to analyze TubeB1 

1. Click to load files

Starting Analysis in just 2 clicks

2. Click to start analysis

Starting Analysis in just 2 clicks

AUTOFLOW

AUTOFLOW improves LPD test efficiency

MAGIC-DR for Acute Myeloid Leukemia Measurable Residual Disease flow

Tube 1&2: Immature markers

Tube 3: Monocytic tube

ACL/AML-MRD panel

MAGIC-DR pipeline

Shopsowitz K. et al., Cytometry B. under recvision

UMAP

Traditional bivariate plots

Traditional bivariate plots vs UMAP plot of a post-induction bone marrow sample

B. L. Wood, Curr Protoc Cytom. 93 (2020).

Dimensional Reduction

Building AML Classifiers

Shopsowitz K. et al., Cytometry B. under recvision

UMAP

Characterization of training data

Shopsowitz K. et al., Cytometry B. under recvision

Benchmarking classifiers performance

Shopsowitz K. et al., Cytometry B. under recvision

AML Classifier identified most useful markers

Shopsowitz K. et al., Cytometry B. under recvision

AML Classifier prediction

Ground Truth

Classifier + UMAP for detecting residual cells

Shopsowitz K. et al., Cytometry B. under recvision

Residual cells with distinct phenotype

Residual cells with NOT SO distinct phenotype

Example 2:  AML MRD+

UMAP

Bivariate plots

MAGIC-DR AML-MRD analysis using flowjo

Example 1:  AML MRD-

Shopsowitz K. et al., Cytometry B. under recvision

Conclusion

• Integrating a supervised classifier with unsupervised dimension reduction offers a robust method for AML MRD analysis that can be seamlessly integrated into conventional workflows.

• Our approach can support and augment human analysis by highlighting abnormal populations that can be gated on for quantification and further assessment.

AUTOFLOW 3.0

Sample-Prep Robot

General Neural Net model for clinical flow labs

Fully automate sample prep steps

Automation

Supps

Group similar datapoints(cell) together into pre-defined number of groups

Objective: Minimize within group difference

How k-means work

Quiz 

Quiz 

?

Quiz 

Number of plots increases as numbers of parameters increase (∑(n-1)^th)

S. C. Bendall et al., Science. 332, 687–696 (2011).

Analysis complexity

Expression Matrix

F. Hahne et al., Bmc Bioinformatics. 10, 106 (2009).

Compensation

Transformation

Filtering

Subsetting

Normalization

...

Data preproccession

L. McInnes, J. Healy, J. Melville, Arxiv (2018). 

https://pair-code.github.io/understanding-umap/

tSNE vs. UMAP

AUTOFLOW vs. Manual

AUTOFLOW vs. Manual