Change the way the main `provoc` function works.

[DBecker7]

The interfaces requires data frames with particular formatting and column names. This is not acceptable.

Ideally, the provoc function should take something like a formula argument similar to a binomial glm, e.g.,

provoc(
    formula = cbind(count_column_name, coverage_column_name) ~ B.1.1.7 + B.1.617.2,
    data = mydata,
    mutation_defs = astronomize(),
    by = "sra"
)

where the user can specify the lineages however they want. If they supply a matrix with column names corresponding to their lineages it should work, otherwise it should use a built-in set of definitions (currently astronomize(), but this can change). The lineage definition will be joined with mydata according to the relevant mutations, potentially giving a warning if there aren't many mutations in common (in practice, there are very few mutations used in lineage definitions, so the warning cutoff may need to be quite low).

If mutation_defs is NULL, then the function should check the column names of mydata.

cbind(count, coverage) ~ . should check for known definition names in mydata and simply use those as the definitions. If ambiguous, return an error.

Note that count_column_name in cbind(count_column_name, coverage_column_name) is interpreted as a column in mydata - it doesn't need to be a vector in the user's environment. This will require some knowledge of parsing formulas in R.

Currently, the function looks for a column labelled "sample", then applies the model to each sample separately. This should be replaced by a by argument, which allows the user to specify a column of their choosing (in this example, mydata$sra). This may be multiple columns, such as location and date, which uniquely define each sample.

Ideally, the output should be an object of class provoc.

[danerkestey]

Hi @DBecker7, thank you for your suggestions. I agree with your improvements, here are some possible improvements to make it more adaptable and user-friendly:

• Implement formula arguments to allow users to specify their count, coverage columns, lineages
• Check for definition names for how we handle lineage definitions, ie. check mydata when mutation_defs is NULL
• Error handling for ambiguous definitions and warnings for mutation overlap
• Replace sample with by argument to let users define samples using their own column names
• Make sure the output of the function is returning an object of class provoc

Let me know what you think about these improvements, and if I should alter my approach