Link Search Menu Expand Document

Great....yet another TMA dearray program. What does this one do?

Coreograph (code) uses UNet (Ronneberger et al., 2015), a deep learning model, to identify complete/incomplete tissue cores on a tissue microarray, and export them individually for faster downstream processing. It has been trained on 9 TMA slides of different sizes and tissue types.

Training sets were acquired at 0.65 microns/pixel resolution and downsampled 1/32 times, or 25 times, to speed up performance. Once the center of each core has been identifed, active contours is used to generate a tissue mask of each core that can aid downstream single cell segmentation. A GPU is not required but will reduce computation time.

Troubleshooting scenarios

NOTE for HMS users: When using Coreograph on O2, the O2tma profile should be used!!

1. Alright, let’s get started!

When using MCMICRO, Coreograph does not require any additional input parameters to run. The DNA channel is assumed to be in the 1st channel.
--core-opts: <leave blank>

As one can see, each core is labelled with a single number implying that each core was found uniquely. Furthermore, each core has a thick white line to indicate the accuracy of segmenting each core. (Future versions will have a colored outlines for better visibility).

2. Well, my DNA channel is not in the 1st channel.

No problem! Specify --channel with the channel that it’s in. This is 0-indexing. So 1st channel is 0. If it’s in the 4th channel,
--core-opts: --channel 3

3. The cores aren’t being found properly.

Coreograph is trained on various core sizes ranging from 500 microns to 2 mm acquired at a pixel size of 0.65 microns per pixel and then downsampled 5 times. If your core size or image resolution are significantly different, you will need to either upsample or downsample a different number of times using --downSampleFactor. See below for examples:

3a) If your pixel size is 0.325 microns per pixel, then your pixel size is double the training data by a factor of 2 (0.65/0.325). You should downsample more times. Use 6 instead of 5.

--core-opts: --downsampleFactor 6

3b) If your pixel size is 1.3 microns per pixel, then your pixel size is half of the training data (0.65/1.3). Instead of downsampling by 5 times (default), you should downsample less. Try 4.

--core-opts: --downsampleFactor 4