Timeline for What are the effects of adding technical/pseudo replicates to the ones to be deconvolved?
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 10, 2022 at 10:06 | history | bounty ended | CommunityBot | ||
| Nov 6, 2022 at 17:33 | vote | accept | Sam | ||
| Nov 6, 2022 at 17:29 | history | edited | EdM | CC BY-SA 4.0 |
in response to comments
|
| Nov 6, 2022 at 16:40 | comment | added | EdM | @Sam yes, in this application the number of linearly-independent columns in $M_{b,\cdot}$ is limiting. For the general matrix equation shown, the number of linearly-independent rows in $B$ could be limiting instead. You might be able to find a solution with a rank-1 expression matrix (just 1 linearly independent column) like you constructed with the pseudo-replicates, but there is no unique solution and what you are fitting with added noise is essentially the noise. I'll add a bit to the answer now that I know you are limiting to 2 cell types. | |
| Nov 6, 2022 at 12:26 | comment | added | Sam | Even a more trivial check : I've seen that it is possible to de-convolute 1 single sample with CibersortX to get its proportions (using a signature matrix of 9 cell-types). Should not this be impossible according to what you say (the limiting step being getting the list of fractions )? Perhaps I mistunderstand something | |
| Nov 6, 2022 at 11:06 | comment | added | Sam | Not sure your explanation is correct.I've created an artificial mixture which consists of one sample and 19 pseudo-replicates (not introducing any noise,just copied the sample values 19 times). I've also created a signature matrix with 9 cell types (by averaging the values of the different cell-types in the signature matrix ). I have managed to run CibersortX and get the cell-types fraction matrix F (not using CibersortX batch correction). The number of linearly independent samples is 1 and the number of cell types is 9, so this should not have been mathematically possible by your explanation. | |
| Nov 6, 2022 at 9:24 | comment | added | Sam | You say The first problem is that you can't solve that equation uniquely for any more cell types c than you have linearly independent columns in Mb,or _rows in B_ The latter part seems incorrect - the number of rows in B is the number of genes (which is not the limiting factor). | |
| Nov 6, 2022 at 8:21 | vote | accept | Sam | ||
| Nov 6, 2022 at 14:46 | |||||
| Nov 4, 2022 at 16:08 | history | answered | EdM | CC BY-SA 4.0 |