Run the singlets through the optimisation and see if they are correctly classified

[jasonserviss]

This might be able to be utilised as additional information in a "normal" run. Since we know what each singlet is, we can use the lowest fraction correctly assigned to a singlet and the highest fraction incorrectly assigned to a singlet as a measurement of the range of "valid" fractions.

[jasonserviss]

Without an uncertainty parameter and weighted means this isn't working well (it may not even work well with an uncertainty parameter and weighted means). The reason is this... In cases where 2 groups of singlets are close to one another (acinar and ductal for example), since we use the mean of the groups in the deconvolution, it can be the case that an individual singlet, when running the deconvolution on the singlets, is closer to the mean of the "other" group than it is to the mean of its "own" group. In theory this could be happening with doublets as well and therefore, the implications should be considered further.