CellProfiler Analyst 2.2.1 Classifier Rules

Hi, I have just recently started on using the classifier in CP Analyst. I managed to obtain some classifier rules but I don’t understand the “IF” statement and the numbers eg what they mean/represent in the rules. Below is an example of the rule that I have:

IF (nuclei_Intensity_MADIntensity_mito > 0.0037465095520019531, [-0.66219538337257688, 0.66219538337257688], [0.75535238846532238, -0.75535238846532238])

I’m hoping someone could help me interpret this. Thanks.

Sue

Hi @Sue1,

These are the rules that has been generated based on the sample chosen for classification. There would be #number of rules that has been specified. In your example above, one of the rule is, it has chosen the cells that have parameter nuclei_Intensity_MADIntensity_mito value greater than 0.003… though I am not sure about the values in the square brackets. Hope this manual might help too.

Regards,
Lakshmi
Fujifilm Wako Automation (Consultant)
www.wakoautomation.com
For CellProfiler training or optimised pipeline write to,
lakshmi.balasubramanian.contractor@fujifilm.com

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS
https://www.wakoautomation.com/products/yokogawa-high-content-imaging

Essentially what that rule breaks into is:

IF (nuclei_Intensity_MADIntensity_mito > 0.0037465095520019531,

For a given cell/nucleus, determine if the MAD of the intensity in the mito channel is > or =< ~0.00375

[-0.66219538337257688, 0.66219538337257688],

If it IS, subtract 0.66 from its “total likelihood score that it belongs in the first bin”, and add 0.66 to its “total likelihood score that it belongs in the second bin”

[0.75535238846532238, -0.75535238846532238])

If it ISN’T, add 0.75 to its “total likelihood score that it belongs in the first bin”, and subtract 0.75 to its “total likelihood score that it belongs in the second bin”

At the end, it will add up the total likelihood scores for each bin, and put it in the bin with the highest score.

With two classes, the numbers for each class are symmetric like that, because there are only two possible bin choices, but imagine you had 3 classes- dim cells, bright triangular cells, and bright round cells:

A shape measurement might not tell you much about whether it belongs in bin1 but will tell you a lot about whether it belongs in bins 2 or 3, so the pattern for a measure that signifies high roundness might give you rule weights of [number_near_0, big_negative_number, big_positive_number], [number_near_0, big positive_number, big_negative_number])

An intensity measurement tells you if something belongs in bin 1 but not how it breaks into bins 2 vs 3, so a measurement of brightness <0.01 might give you rule weights of [big_positive_number, big_negative_number, big_negative_number], [big_negative_number, middleish_positive_number, middleish_positive_number])

4 Likes