Overview of using apps

There are 3 top-level functions that provide the major means for identifying what apps are installed, what an app can do and for getting an app to use it. These functions are:

Two other crucial concepts concern data stores and tracking failures.

Types of apps

There are 3 types of apps:

  1. loaders (by convention, names starts with load_<data type>)

  2. writers (by convention, names starts with write_<data type>)

  3. generic (no naming convention)

As their names imply, loaders load, writers write and generic apps do other operations on data.

Composability

Most cogent3 apps are “composable”, meaning that multiple apps can be combined into a single function by addition. For example, say we have an app (fit_model) that performs a molecular evolutionary analysis on an alignment, and another app (extract_stats) that gets the statistics from the result. We could perform these steps sequentially as follows

fitted = fit_model(alignment)
stats = extract_stats(fitted)

Composability allows us to simplify this as follows

app = fit_model + extract_stats
stats = app(fitted)

We can have many more apps in a composed function than just the two shown here.

Composability rules

There are rules around app composition, starting with app types. Loaders and writers are special cases. If included, a loader must always be first, e.g.

app = a_loader + a_generic

If included, a writer must always be last, e.g.

app = a_generic + a_writer

Changing the order for either of the above will result in a TypeError.

The next constraint on app composition are the input and output types of the apps involved. Specifically, apps define the type of input they work on and the type of output they produce. For two apps to be composed, the output (or return) type of app on the left (e.g. a_loader) must overlap with the input type of the app on the right (e.g. a_generic). If they don’t match, a TypeError is raised.

An example

I illustrate the general approach for a simple example – extracting third codon positions. As I’m defining a writer, I also need to define the destination (a directory in this case) where it will write to.

Using apps sequentially like functions

The resulting alignment just3rd will be written into the out_dstore directory in fasta format with the same filename as the original data ("primate_brca1.fasta").

Note

m is a DataMember (described here).

Composing a multi-step process from several apps

We can make this simpler by creating a single composed function.

Applying a process to multiple data records

We use a data store to identify all data files in a directory that we want to analyse. process can be then applied to all records in the data store without having to loop.

Note

result is out_dstore.

Other important features

The settings and data analysed will be logged

A log file will be written into the same data store as the output. The log includes information on the conditions under which the analysis was run and fingerprint all input and output files.

Failures are recorded

Any “failures” (see The NotCompleted object) are saved. The data store class provides methods for interrogating those. First, a general summary of the output data store indicates we have 6 records that did not complete.

These occur for this example primarily because some of the files contain sequences that are not aligned

You can track progress

You can do parallel computation

result = process.apply_to(dstore, parallel=True)

By default, this will use all available processors on your machine. (See Parallel computations for more details plus how to take advantage of multiple machines using MPI.)

All of the above

process.apply_to(dstore, parallel=True, show_progress=True)