Zackular 2014

[naarkhoo]

I am not sure, if here is a right platform to ask this question. But I was wonder, how much the data from your lab (Zackular, et al 2014) is comparable to this cohort; Thanks again for making your papers and codes open.

[pschloss]

This is a great platform to ask this type of question! The Zackular data are a subset of the Baxter dataset.

[naarkhoo]

May I ask, if you have tried too pool these two data sets or use one as an external validation set to test the performance of the predictor model ?

[pschloss]

The Baxter dataset includes the Zackular dataset. Since the Zackular dataset was something of a random subset from the larger dataset and had about 20% of the samples, we opted for the leave-one-out cross validation. We felt that using 80% of the data to train on a single random subset containing 20% of the data would have had a negative bias on the evaluation of the model.

[naarkhoo]

I have made a PCoA of these two, trying to explain, what could be the source of variation. May I ask if those samples are re-sequenced, or it was two different sequencing runs ?

Because, I don't know which samples in the Baxter datasets are the Zackular samples - I was hoping the PCoA plot, could help me to see this overlap.

[pschloss]

Everything was resequenced

[nbaxter13]

The overlapping samples in the Baxter dataset are from the same patients' stool samples, but they were different aliquots of those stool samples. So they were re-extracted and re-sequenced.

[naarkhoo]

It is an interesting data set by its own and shows which predictive taxa potentially is sensitive and alters.

[pschloss]

If you want the Zackular data on it's own, you can download the data and metadata from here: http://mothur.org/MicrobiomeBiomarkerCRC/