detect need to transpose input

[ngiangre]

It should be defined whether input/masks need to be transposed - it shouldn't be left up to the user. It's not intuitive what the transpose flag should be...

[AndrewC160]

Should we just establish "Input metadata tables should be oriented with one patient per column. In the event that a transposed table is to be used, the argument 'transpose_input_table' should be used?"

Otherwise detecting the orientation isn't super obvious...it could be a long format, or it could be a dataset with lots of fields.

[ngiangre]

I like having the input metadata conforming to a standard but we would need checks for the format. But if we can do a check and transpose then we don’t need the argument, just saying that the format should be either way. Does that make sense? I like having less arguments.

On Wed, Jan 20, 2021 at 8:07 PM Andrew Clugston ***@***.***> wrote: Should we just establish "Input metadata tables should be oriented with one patient per column. In the event that a transposed table is to be used, the argument 'transpose_input_table' should be used?" — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#25 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3TDSZSRLVMXSVS7HWHXETS2543VANCNFSM4VR47QCA> .

-- Cheers, *Nick Giangreco* Columbia University Systems Biology Ph.D. Student 716.713.2124

[AndrewC160]

I agree, but what would we use to decide the orientation? If the user provides a CSV/TSV without knowable column/row names we'd need to figure out which are sample names and which are metadata values. If the dataset is sufficeintly large we could have a check for length/width, i.e. "Either they collected 30,000 metadata values or this table is long", but that's not a guarantee...especially with test data that we use and that people will probably feed in.

[ngiangre]

You’re right. So maybe we should keep the parameter but be explicit on the default format?

On Wed, Jan 20, 2021 at 8:34 PM Andrew Clugston ***@***.***> wrote: I agree, but what would we use to decide the orientation? If the user provides a CSV/TSV without knowable column/row names we'd need to figure out which are sample names and which are metadata values. If the dataset is sufficeintly large we could have a check for length/width, i.e. "Either they collected 30,000 metadata values or this table is long", but that's not a guarantee...especially with test data that we use and that people will probably feed in. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#25 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3TDS5UTG5QGT2TG6EWIBTS26AEDANCNFSM4VR47QCA> .

-- Cheers, *Nick Giangreco* Columbia University Systems Biology Ph.D. Student 716.713.2124

[AndrewC160]

Another option is to have the package figure it out when the user provides a mask ("Does row 1 or column 1 have more of the following aliases?"), but I think that would require a bit of reconfiguring of the work flow.

[ngiangre]

Let’s do the path of least resistance first!

On Wed, Jan 20, 2021 at 8:41 PM Andrew Clugston ***@***.***> wrote: Another option is to have the package figure it out when the user provides a mask ("Does row 1 or column 1 have more of the following aliases?"), but I think that would require a bit of reconfiguring of the work flow. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#25 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3TDS3Q25XL5TDMZS5TRDLS26A3FANCNFSM4VR47QCA> .

-- Cheers, *Nick Giangreco* Columbia University Systems Biology Ph.D. Student 716.713.2124