1 of 29

BroadHacks 2024 Workshop – �Imaging Data Analysis with Morpheus

​

Esteban Miglietta

Postdoctoral Associate

Beth Cimini’s lab

Imaging Platform

Broad Institute of MIT and Harvard

2 of 29

Outline

  • Overview of Image-based Profiling & Morpheus
  • Demonstration
  • Hands-on exercise

3 of 29

Outline

  • Overview of Image-based Profiling & Morpheus
  • Demonstration
  • Hands-on exercise

4 of 29

Digital images are ultimately just arrays

​

​

​

5 of 29

Quantitative microscopy

In a typical quantitative microscopy experiment, the protein of interest or a biological structure is labelled with an antibody or dye and one or a few features are measured.

​

Created with BioRender.com

6 of 29

Image-based profiling

In contrast, in image-based profiling, we let the ‘cells speak for themselves’

Created with BioRender.com

7 of 29

Image-based profiling workflow

​

Chandrasekaran et. al NRDD 2020

Cell Painting

6 stains, 5 channels imaged, revealing 8 constituents/organelles:

Nucleus

ER

Nucleoli,

Cytoplasmic RNA

Actin,

Golgi apparatus,

Plasma Membrane

Mitochondria

?

8 of 29

In practice, how do we look at profiles?

Similarity matrices in Morpheus (https://software.broadinstitute.org/morpheus/)

9 of 29

Morpheus

  • Free web-based software - to explore the overall large-scale associations of the data.
  • Originally designed at the Broad Institute for exploration of mRNA profiling data but accepts a variety of matrix files from multiple formats (CSV, GCT, GMT, text file) to be imported.
  • Allows matrix visualization, analysis, clustering, filtering and displaying of charts.
  • No extensive computational or statistical experience is required.
  • Helps to gain insights into the biological interpretation of the profiles.

​

​

10 of 29

Applications:

​

  • To examine correlations between replicates to check for their variability. 🡪 QC!
  • To examine correlations among the perturbations (i.e. drugs with known and unknown MOA).
  • To discern what features drive differences between samples or groups (Marker selection).
  • To interpret the biology behind the data.

11 of 29

Workflow - Morpheus

Created with BioRender.com

12 of 29

Workflow - Morpheus

13 of 29

Extraction of features

  • CellProfiler

​

  • Can be done with any image analysis software

14 of 29

Workflow - Morpheus

15 of 29

Processing

  • Average per well
    • Easier to compare ~96/384 wells than comparing millions of cells.
    • Lot of heterogeneity in cells will make clustering tougher.
    • With this approach we can look at major systemic changes and not the subtle ones.

​

  • Normalization
    • Either to all data or to the negative controls
    • Why it is important?

​

  • Select a subset of features
    • Lot of features with the same information will be used in an additive way in downstream analysis which will not reflect the actual biology.
    • Remove redundant features and outliers, balance weights of features.

​

16 of 29

Workflow - Morpheus

17 of 29

Pearson correlation coefficient

​

  • A Pearson correlation coefficient is a way of representing the measurement of similarity, where it measures the strength of the linear relationship between two variables (in our case, between two wells across a large set of features or between two features across a large set of wells).

​

  • Pearson coefficient:
    • 1 🡪 perfect positive correlation
    • 0 🡪 no correlation
    • –1 🡪 perfect negative correlation.

Profile of well B2

Profile of well C2

18 of 29

Workflow - Morpheus

19 of 29

Similarity matrix

​

  • A similarity matrix is a way to assess the covariance in features between all pairs of columns or rows.
  • In each square of the matrix, a Pearson correlation coefficient was calculated for all features in the dataset between each pair of samples.
  • The squares at the intersection of those two samples are set as the value of that correlation coefficient, and so on for each pair of wells.

-1

1

0

20 of 29

Similarity matrix

​

  • A similarity matrix is a way to assess the covariance in features between all pairs of columns or rows.
  • In each square of the matrix, a Pearson correlation coefficient was calculated for all features in the dataset between each pair of samples.
  • The squares at the intersection of those two samples are set as the value of that correlation coefficient, and so on for each pair of wells.

-1

1

0

21 of 29

Caveats:

​

  • For these analyses to be valid at all, you need to be sure that
    • Your segmentation is “good enough” across all your conditions
    • See blog post for guidelines

​

  • Doing this level of comparison is typically a starting point, not an ending point
    • LOOKING at your data (both at the image stage and the bioinformatic stage) is critical
    • Consider the “Datasaurus Dozen” (Justin Matejka and George Fitzmaurice)

​

22 of 29

Conclusions:

  • Morphological profiling can be a powerful way to explore cellular phenotype data
  • Measure lots of things in CellProfiler or image analysis software – you never know what may end up being useful!
  • By collapsing, normalizing, and feature selecting our data, we can turn information about millions of cells into something we can easily explore for biological insight
  • Morpheus helps to analyze the image-based profiles without the use of command line

23 of 29

Outline

  • Overview of Morpheus
  • Demonstration - software.broadinstitute.org/morpheus/
  • Hands-on exercise

24 of 29

Demonstration:

  • Navigation
  • Normalization - robust z-score
  • Filtering of features
  • Pearson correlation coefficient
  • Similarity matrix
  • Hierarchical clustering
  • Marker/feature selection

​

25 of 29

Outline

  • Overview of Morpheus
  • Demonstration
  • Hands-on exercise

26 of 29

Details on the exercise

  • Dataset:
    • We will be using BBBC021 data set, a dataset from the Broad Bioimage Benchmark Collection.
    • Images of MCF7 cancer cells
      • Treated with 113 compounds at 8 concentrations
      • Stained for stains for DNA, actin, and tubulin.
      • Negative control = DMSO
  • Aim:
    • We want to know how each well is similar to each other well using similarity matrix
  • Team:
    • Suganya Sivagurunathan, Postdoctoral Associate
    • Paula Llanos, Postdoctoral Associate
    • Shatavisha Dasgupta, Postdoctoral Associate

27 of 29

Acknowledgement:

​

28 of 29

Resources:

​

​

29 of 29