Different paired fc result in MetaboAnalyst and excel

[linxiaotong]

Hi, thanks so much for developing such a helpful online site; I am unclear about the fold change calculation for paired data and would like to ask you for some advice.

In excel， I used the day01 concentration of metabolite A for each subject, divided by the day28 concentration to get the result in the foldchange column and found the mean value of these 9 fold changes to be 225.1389472.

However, when I put the same data into the MetaboAnalyst website (including all metabolites, not just A_unsure), I get a different calculation, with the FC result for A_unsure being 7.3864 and log2(FC) being 2.88487.

I am baffled as to what the problem is and am looking forward to your help!

here mydata

Sample | paried_group | A_unsure | (01/28)foldchange | mean_foldchange
people0101 | -1 | 0.004127571 | 0.031645023 | 225.1389472
people0201 | -2 | 0.83822522 | 34.23023572 |
people0301 | -3 | 0.106605361 | 2.410679456 |
people0401 | -4 | 0.631875643 | 2.907502789 |
people0501 | -5 | 0.572538855 | 835.1818239 |
people0601 | -6 | 0.904662272 | 1.369159927 |
people0701 | -7 | 0.842034249 | 5.33574297 |
people0801 | -8 | 0.063300182 | 1.235493619 |
people1001 | -9 | 0.792196475 | 1143.548242 |
people0128 | 1 | 0.130433495 | |
people0228 | 2 | 0.02448786 | |
people0328 | 3 | 0.044222122 | |
people0428 | 4 | 0.217325894 | |
people0528 | 5 | 0.000685526 | |
people0628 | 6 | 0.660742587 | |
people0728 | 7 | 0.157810122 | |
people0828 | 8 | 0.05123473 | |
people1028 | 9 | 0.000692753 | |

and in web ,the paired fc result is 7.3864

[jeff.xia]

There are data processing steps including filtering, missing value estimation, normalization involved. This could have big impact on your FC computing especially for those features containing missing or very small values. You can refer to “data_processed.csv” to find whether this is the case for your features

[linxiaotong]

Thanks for your reply~
I checked all three files(data_normalized,data_original,data_processed) and they are consistent.
Download R-history that my data has no missing values and is not filtered and normalised (as the demo is from the normalised file)

thank u~

[jeff.xia]

Have you read our FC analysis for paired data? You can also take a look at our MetaboAnalystR code.

[linxiaotong]

Thanks for your advice, I will take a look at codes in r