Example Workflow with spatialproteomics

Welcome to spatialproteomics! In this notebook, we will go through an example workflow by looking at the following steps: 1. Reading in a highly multiplexed image and creating a spatialproteomics object 2. Performing basic image processing steps to boost the signal-to-noise ratio 3. Performing cell segmentation using cellpose 4. Quantifying protein expression per cell 5. Predicting cell types with a simple argmax technique 6. Plotting the results

[1]:

%load_ext autoreload
%autoreload 2

import spatialproteomics as sp
from skimage.io import imread
import matplotlib.pyplot as plt
from scipy.signal import medfilt2d

2. Image Processing

Highly multiplexed fluorescence imaging techniques frequently suffer from poor signal-to-noise ratio. To alleviate this problem, you can threshold out low intensity pixels, thereby boosting the contrast of the image. While there are automated methods to determine the thresholds for such operations, it is difficult to come up with one that works in all cases. Here, we therefore set the thresholds based on manual inspection.

Let’s start by looking at the data set in the current form.

[5]:

_ = sp_object.pl.show()  # pl: the plotting module

../_images/notebooks_ExampleWorkflow_10_0.png

This is a bit much to look at. First, we want to check only the DAPI channel, to get a feel for the signal-to-noise ratio.

[6]:

_ = sp_object.pp['DAPI'].pl.show()  # we can use pp[channels] to subset the spatialproteomics object

../_images/notebooks_ExampleWorkflow_12_0.png

This looks promising. What about the other channels?

[7]:

_ = sp_object.pp[['CD4', 'CD8', 'FOXP3', 'BCL6']].pl.show() # pp: the preprocessing module

../_images/notebooks_ExampleWorkflow_14_0.png

Looking good, but we can make the signal a bit clearer by performing some image processing. In the following codeblock, we first threshold the channels by some percentile (so every value below that percentile gets set to 0). We then apply a 2D median filter with a kernel size of 3 to apply some smoothing.

[8]:

# the percentiles by which the channels will be thresholded
percentiles = [0.2, 0.6, 0.6, 0.6, 0.6]

# thresholding and smoothing the data
sp_object = sp_object.pp.threshold(percentiles).pp.apply(medfilt2d, kernel_size=3)

Let’s plot the new object.

[9]:

fig, ax = plt.subplots(1, 2, figsize=(10, 5))
_ = sp_object.pp['DAPI'].pl.show(ax=ax[0])
_ = sp_object.pp[['CD4', 'CD8', 'FOXP3', 'BCL6']].pl.show(ax=ax[1])

../_images/notebooks_ExampleWorkflow_18_0.png

6. Plotting

Finally, let’s do some plotting of the predicted cell types next to the markers. Before plotting, we can set some colors for the cell types (labels).

[21]:

# setting the colors for the cell types
sp_object = sp_object.la.set_label_colors(['T_h', 'T_tox', 'T_reg', 'T_fh'], ['red', 'green', 'yellow', 'blue'])

[22]:

fig, ax = plt.subplots(1, 3, figsize=(15, 5))
_ = sp_object.pp['DAPI'].pl.show(ax=ax[0])
_ = sp_object.pp[['CD4', 'CD8', 'FOXP3', 'BCL6']].pl.show(ax=ax[1])
_ = sp_object.pl.show(render_image=False, render_segmentation=True, render_labels=True, ax=ax[2])

../_images/notebooks_ExampleWorkflow_39_0.png

And this is how easy it can be to perform analysis of highly multiplexed immunofluorescence images! If you have any additional questions, check out the other notebooks for details.

Example Workflow with spatialproteomics

1. Getting Started

2. Image Processing

3. Cell Segmentation

4. Quantifying Protein Expression per Cell

5. Cell Type Prediction

6. Plotting