Vala Sciences

CyteSeer calculates Manders’ M1 and M2 from channel 1 vs. channel 2 in the colocalization (formerly the “perilipin”) algorithm. The Manders’ coefficients were publisedh by E. M. M. Manders, et al. 1993, “Measurement of co-localization of objects in dual-colour confocal images”. Journal of Microscopy 169:375-382.

Shown, here, is a representation of the original equations from the publication. Note that Li is the “lipid image”, and Pi is the “Protein image”. the subscript “coloc” refers, in the original publication, to pixels that are > 0 in intensity in the second channel. In other words, pixels that are “colocalized” would be positive in the lipid image, and also positive in the protein image.

Thus, M1 is the sum of intensities in the lipid channel of all of the pixels that are at locations that are also >0 in the protein channel, divided by the sum of intensities for all of the pixels in the lipid image. M2 is the same calculation, but using the protein image as the reference image.

The problem is that in “real life” in the images collected by digital cameras and automated microscopy, pixel intensities are always > 0 at every pixel location in every channel. Thus, M1 and M2 will always resolve to exactly 1.0, which doesn’t convey any information, at all.

CyteSeer uses two separate versions of the Manders’ M1 and M2 correlation coefficients that return meaningful values regarding colocalization.

The first version of the Manders’ coefficients which are the data parameter identified as “M1” and “M2” in CS’s data table, were derived as as follows:

1) CS performs the segmentation, utilizing the Lipid Image to derive the Lipid Mask (Lm), this is the mask that represents the lipid droplets, and the Protein Image to drive the Protein Mask (Pm) this is the mask that represents the protein (such as perilipin or HSL).

2) For the numerator, the “coloc” pixels for M1 are the pixels in the Lipid Image that are in the Pm derived from the protein image. For M2, the “coloc” pixels are the pixels in the Protein Image that are in the Lm derived from the Lipid Image.

The second version of the Manders’ coefficients are the “Masked Manders’” coefficients, which are the parameters identified as “M M1” and “M M2” in CS’s data table:

1) CS performs the segmentation, utilizing the Lipid Image to derive the Lipid Mask (Lm), this is the mask that represents the lipid droplets, and the Protein Image to drive the Protein Mask (Pm) this is the mask that represents the protein (such as perilipin or HSL).

2) For the Lipid Image, all pixels that are NOT part of Lm are assigned a pixel intensity value of 0. For the Protein Image, all pixels that are NOT part of the Pm are assigned a pixel intensity value of 0. Then the same calculation is performed.

For Masked M1, only pixels that are part of BOTH the Lm and Pm will contribute to the numerator of M1 since non Lm pixels in the Lipid Image have an intensity value of 0 from step #1). Note that this was not true of the unmasked version of M1 pixels in the Pm for the Lipid Image are not necessarily within the Lm.

Thus, the M M1 and M M2 values are more “specific”, than the unmasked M1 and M2 values (or the M1 and M2 relationships defined by Randy on the previous page), as