Short term change

[Daenarys8]

Function to return short term change in a dataframe

[Daenarys8]

@antagomir @TuomasBorman

[TuomasBorman]

In holiday currently. I will check this early next week.

[TuomasBorman]

This PR is probably also not the most urgent ones

The calcultion part / core functionality is rather simple. The function should:

1. Get data with mia::meltSE (transformations sould be applied in prior to this function)
2. Calculate short term change based on group and time.col
3. Return the data.frame

After this, user could plot the results if wanted

[antagomir]

Good to finalize this as proposed.

[antagomir]

Example shows the following:

# Load time series data
#' data(minimalgut)
#' tse <- minimalgut
#' 
#' short_time_labels <- c("74.5h", "173h", "438h", "434h", "390h")
#' 
#' # Subset samples by Time_lable and StudyIdentifier
#' tse <- tse[, !(colData(tse)$Time_label %in% short_time_labels)]
#' tse <- tse[, (colData(tse)$StudyIdentifier == "Bioreactor A")]
#' 
#' # Plot short term change in abundance
#' shortTermChange(tse, rarefy = TRUE, plot = TRUE) + ggtitle("Bioreactor A")

Is this mean to be an analysis or visualization function?

The example could explain a bit more what this aims to show. I am not sure if I understand the motivation of this function from the manpage description.

It would be good to have a separate function, like getShortTermChange / addShortTermChange and then a separate function for visualization (unless it is easy to do with existing functions, like miaViz::plotSeries).

Is this linked to some miaTime issue? Issue #7 ?

[Daenarys8]

#'
#' # Plot short term change in abundance
#' shortTermChange(tse, rarefy = TRUE, plot = TRUE) + ggtitle("Bioreactor A")

Is this mean to be an analysis or visualization function?

original reference had plotting as part of the function. It is now removed and the function returns a dataframe with short term changes

Is this linked to some miaTime issue? Issue #7 ?

yes, syncomR consumed phyloseq object.

[Daenarys8]

Moreover given the nature of the data frame results, I am not sure plotSeries is the right visualization method for it. Suppose we somehow addShortTermChange using one of the tse slots(say maybe reducedDims), it would still require it's own ggplot wrapper .
I just checked but to be sure @TuomasBorman do we have miaViz function that takes dataframes?
For reference, this is how the plots from syncomR could look like linear or polarized:

and of course we can also discuss how we want to represent our own plots.

[TuomasBorman]

Looks nice

[TuomasBorman]

Moreover given the nature of the data frame results, I am not sure plotSeries is the right visualization method for it. Suppose we somehow addShortTermChange using one of the tse slots(say maybe reducedDims), it would still require it's own ggplot wrapper . I just checked but to be sure @TuomasBorman do we have miaViz function that takes dataframes? For reference, this is how the plots from syncomR could look like linear or polarized:

and of course we can also discuss how we want to represent our own plots.

There is no that kind of restriction that miaViz cannot have functions that take df as input. In fact, loadings plotting function can take df as input,

This outputs currently data.frame and I think it makes sense. reducedDims is not correct because this does not reduce dimensionality. This could be stored to metadata and then we could have a plotting function for TreeSE and df.

[antagomir]

Or can it be part of colData? Could be simpler. This is a measure that is available on a per-sample basis. Except that some samples may have NAs if they are missing the comparison.

[TuomasBorman]

Or can it be part of colData? Could be simpler. This is a measure that is available on a per-sample basis. Except that some samples may have NAs if they are missing the comparison.

That could be possible but then columns would denote features. It might be confusing and there might be lots of features to add in the columns

[antagomir]

Or can it be part of colData? Could be simpler. This is a measure that is available on a per-sample basis. Except that some samples may have NAs if they are missing the comparison.

That could be possible but then columns would denote features. It might be confusing and there might be lots of features to add in the columns

Yep, not so optimal.

[antagomir]

"This function essentially calculates the change in abundance over time for a microbe by abundancet2/ abundancet1."

We already have getStepwiseDivergence?

getStepwiseDivergence calculates x(t + d) - x(t) for abundance x at time points t+d vs. t any user defined interval d along the time series.

getShortTermChange calculates this per taxa. One can generalize and use it for a more general interval x(t+d)/x(t), analogous to getStepwiseDivergence.

getShortTermChange suggests to do x(t+d)/x(t) instead of x(t+d)-x(t). But both are available and the choice may also depend on whether the data is logarithmic. log(a/b) = log(a)-log(b) although zeroes may cause trouble. It could be enough to implement the difference "x-y" and the rest ("x/y") can be always dealt (by the user?) with log/exp operations the data if necessary.

[TuomasBorman]

Looks good, some suggestions

[TuomasBorman]

You could use math notation, it would be easier to read. Also the calculated values include other measurements than this

[TuomasBorman]

You could describe them also

[TuomasBorman]

The values should be also checked

[antagomir]

I somehow missed what was the motivation for this function?

[TuomasBorman]

I somehow missed what was the motivation for this function?

Related to this issue: #7