Matt:LabNotes/2015-11-6
Jump to navigation
Jump to search
Analyzing RNAscope Images[edit]
- Genes
- PDE1A (expect in layers 2,4, and 5)
- SLC17A7 (expect everywhere but not uniform)
- OLFM1
- RELN (expect in layer 1)
- Three samples
- PDE1A + OLFM1 + SLC17A7
- PDE1A + OLFM1
- RELN + SLC17A7
Strategy[edit]
- Images are RGB so need to convert to grayscale
- Green channel can just use green RGB values
- Magenta channel has equal Red and Blue RGB values at every pixel so just use blue RGB values
- Yellow channel has a lot of red background and even though green RGB values are lower at spots it is high enough; use green RGB values
- Divide each tile (which is approx 5,600 x 49,660 pixels) into 48 2048x2048 sub-images
- With long edge aligned vertically, cut horizontally into 24 sub-images every 2048 pixels
- Cut vertically the middle 4096 pixels into two 2048 strips (indices 786-2833 and 2834-4881)
- Count number of nuclei and spots for each gene in each sub-image
- Expect to see some pattern or trend in vertical direction
- Left and right sub-images are like replicate samples and should be similar
- Images are RGB so need to convert to grayscale
Counting transcript dots[edit]
- Apply gaussian filter
- Tried sigma = 0.5, 0.75, 0.8, and 1 and found 0.75 works best
- Apply threshold
- Tried 10 and 20 and found 20 to be best (avoids false positives from next step)
- Find local maximum (peaks)
- Count with bwconncomp
Validate[edit]
- c3_3-1 cropped (increased and brightness and contrast) -> marked local max after filtering
File:C3 3-1 cropped.jpg -> File:C3 3-1 cropped gaussfilt75.jpg
- c4_1-2 cropped -> marked local maximum after filtering
File:C4 1-2 cropped.jpg -> File:C4 1-2 cropped gaussfilt75.jpg
Counting nuclei[edit]
- nucC(Image,40)
- Erode and dilate to remove noise within nuclei
- Strel size = 40
- Convert to binary image
- Segment with watershed transformation
- NucC.m counts every "little" nuclei while another script: count_nuclei_63x.m counts only large "typical" nuclei
- Using NucC since I think it more accurately represents number of cells in the image
Results[edit]
- Divided transcript counts by number of nuclei in each 2048x2048 image
- Calculated average and standard error of the two replicates (left and right images)
- Plotted average and used error bars to represent SEM
RELN+SLC17A7[edit]
- Tile 1
- RELN has expected spike
- Actually only a couple cells had all the RELN expression, not uniform in that image layer
File:R+S Tile 1 Spots Per Nuclei.JPG
- Tile 2
- Y-slices 20-24 had very few or no cells because tissue ended there so removed them
File:R+S Tile 2 Spots Per Nuclei.JPG File:R+S Tile 2 Spots Per Nuclei RELNzoomed.JPG
PDE1A+OLFM1[edit]
- Yslice 24 had no nuclei
- Yslices 20-24 had very little nuclei and nuclei counting algorithm tended to overestimate for these images, sometimes by >100%
- I notice similarities in spikes at Y-Slice 5, 10, 15, and 18-19
- Tile 1
File:P+O Tile 1 Spots Per Nuclei.JPG File:P+O Tile 1 Spots Per Nuclei PDE1Azoomed.JPG
- Tile 2
File:P+O Tile 2 Spots Per Nuclei.JPG File:P+O Tile 2 Spots Per Nuclei PDE1Azoomed.JPG
PDE1A+OLFM1+SLC17A7[edit]
- Tile 1
File:P+S+O Tile 1 Spots Per Nuclei.JPG File:P+S+O Tile 1 Spots Per Nuclei zoomedPDE1A.JPG
- Tile 2
File:P+S+O Tile 2 Spots Per Nuclei.JPG File:P+S+O Tile 2 Spots Per Nuclei zoomedPDE1A.JPG