How to add a new report

In this tutorial, we will show how to add a new report to plot sequence length distribution in repertoire datasets. This tutorial assumes you have installed immuneML for development as described at Set up immuneML for development.

Adding a new Report class

To add a new report, add a new class to the data_reports package. In this example, the new class is called NewSequenceLengthDistribution.

The NewSequenceLengthDistribution class should inherit DataReport class and implement all abstract methods.

An example implementation is shown below. It includes implementations of abstract methods build_object(**kwargs), check_prerequisites() and _generate(), and class documentation at the beginning. This class documentation will be shown to the user.

import logging
from collections import Counter
from pathlib import Path

import pandas as pd
import plotly.express as px

from immuneML.data_model.dataset.RepertoireDataset import RepertoireDataset
from immuneML.data_model.repertoire.Repertoire import Repertoire
from immuneML.reports.ReportOutput import ReportOutput
from immuneML.reports.ReportResult import ReportResult
from immuneML.reports.data_reports.DataReport import DataReport
from immuneML.util.PathBuilder import PathBuilder


class NewSequenceLengthDistribution(DataReport):
    """
    Generates a histogram of the lengths of the sequences in a RepertoireDataset.

    YAML specification:

    .. indent with spaces
    .. code-block:: yaml

        my_sld_report: NewSequenceLengthDistribution

    """

    @classmethod
    def build_object(cls, **kwargs): # called when parsing YAML - all checks for parameters (if any) should be in this function
        return NewSequenceLengthDistribution(**kwargs)

    def __init__(self, dataset: RepertoireDataset = None, batch_size: int = 1, result_path: Path = None, name: str = None):
        super().__init__(dataset=dataset, result_path=result_path, name=name)
        self.batch_size = batch_size

    def check_prerequisites(self): # called at runtime to check if the report can be run with params assigned at runtime (e.g., dataset is set at runtime)
        if isinstance(self.dataset, RepertoireDataset):
            return True
        else:
            logging.warning("NewSequenceLengthDistribution: report can be generated only from RepertoireDataset. Skipping this report...")
            return False

    def _generate(self) -> ReportResult: # the function that creates the report
        sequence_lengths = self._get_sequence_lengths()
        report_output_fig = self._plot(sequence_lengths=sequence_lengths)
        output_figures = None if report_output_fig is None else [report_output_fig]
        return ReportResult(type(self).__name__, output_figures=output_figures)

    def _get_sequence_lengths(self) -> Counter: # implementation detail: extract sequence lengths from repertoires in the dataset
        sequence_lenghts = Counter()

        for repertoire in self.dataset.get_data(self.batch_size):
            seq_lengths = self._count_in_repertoire(repertoire)
            sequence_lenghts += seq_lengths

        return sequence_lenghts

    def _count_in_repertoire(self, repertoire: Repertoire) -> Counter: # implementation detail: get lengths of sequences for one repertoire
        c = Counter([len(sequence.get_sequence()) for sequence in repertoire.sequences])
        return c

    def _plot(self, sequence_lengths: Counter) -> ReportOutput: # implementation detail: when all lengths are know, plot them

        df = pd.DataFrame({"counts": list(sequence_lengths.values()), 'sequence_lengths': list(sequence_lengths.keys())})

        figure = px.bar(df, x="sequence_lengths", y="counts")
        figure.update_layout(xaxis=dict(tickmode='array', tickvals=df["sequence_lengths"]), yaxis=dict(tickmode='array', tickvals=df["counts"]),
                             title="Sequence length distribution", template="plotly_white")
        figure.update_traces(marker_color=px.colors.diverging.Tealrose[0])
        PathBuilder.build(self.result_path)

        file_path = self.result_path / "sequence_length_distribution.html"
        figure.write_html(str(file_path))
        return ReportOutput(path=file_path, name="sequence length distribution plot")

Unit testing the new Report

To add a unit test:

  1. Add a new file to data_reports package named test_newSequenceLengthDistribution.py.

  2. Add a class TestNewSequenceLengthDistribution that inherits unittest.TestCase to the new file.

  3. Add a function setUp() to set up cache used for testing (see example below). This will ensure that the cache location will be set to EnvironmentSettings.tmp_test_path / "cache/"

  4. Define one or more tests for the class and functions you implemented.

  5. If you need to write data to a path (for example test datasets or results), use the following location: EnvironmentSettings.tmp_test_path / "some_unique_foldername"

Typically, the generate_report() function of the new report should be tested, as well as other relevant methods, to ensure that the report output is correct. When building unit tests, a useful class is RandomDatasetGenerator, which can create a dataset with random sequences.

An example of the unit test TestNewSequenceLengthDistribution is given below.

import os
import shutil
from unittest import TestCase

from immuneML.caching.CacheType import CacheType
from immuneML.data_model.dataset.RepertoireDataset import RepertoireDataset
from immuneML.data_model.receptor.receptor_sequence.ReceptorSequence import ReceptorSequence
from immuneML.data_model.repertoire.Repertoire import Repertoire
from immuneML.environment.Constants import Constants
from immuneML.environment.EnvironmentSettings import EnvironmentSettings
from immuneML.reports.data_reports.NewSequenceLengthDistribution import NewSequenceLengthDistribution
from immuneML.util.PathBuilder import PathBuilder


class TestNewSequenceLengthDistribution(TestCase):

    def setUp(self) -> None:
        os.environ[Constants.CACHE_TYPE] = CacheType.TEST.name

    def test_generate_report(self):
        path = EnvironmentSettings.tmp_test_path / "datareports"
        PathBuilder.build(path)

        rep1 = Repertoire.build_from_sequence_objects(sequence_objects=[ReceptorSequence(amino_acid_sequence="AAA", identifier="1"),
                                                                        ReceptorSequence(amino_acid_sequence="AAAA", identifier="2"),
                                                                        ReceptorSequence(amino_acid_sequence="AAAAA", identifier="3"),
                                                                        ReceptorSequence(amino_acid_sequence="AAA", identifier="4")],
                                                      path=path, metadata={})
        rep2 = Repertoire.build_from_sequence_objects(sequence_objects=[ReceptorSequence(amino_acid_sequence="AAA", identifier="5"),
                                                                        ReceptorSequence(amino_acid_sequence="AAAA", identifier="6"),
                                                                        ReceptorSequence(amino_acid_sequence="AAAA", identifier="7"),
                                                                        ReceptorSequence(amino_acid_sequence="AAA", identifier="8")],
                                                      path=path, metadata={})

        dataset = RepertoireDataset(repertoires=[rep1, rep2])

        report = NewSequenceLengthDistribution(dataset, 1, path)

        result = report.generate_report()
        self.assertTrue(os.path.isfile(result.output_figures[0].path))

        shutil.rmtree(path)

Adding a Report: additional information

In immuneML, it is possible to automatically generate a report describing some aspect of the problem being examined. There are a few types of reports:

  1. Data report – reports examining some aspect of the dataset (such as sequence length distribution, gene usage)

  2. Encoding report – shows some aspect of the encoded dataset (such as the feature values of an encoded dataset),

  3. ML model report – shows the characteristics of an inferred machine learning model (such as coefficient values for logistic regression or kernel visualization for CNN)

  4. Train ML model report – show statistics of multiple trained ML models in the TrainMLModelInstruction (such as comparing performance statistics between models, or performance w.r.t. an encoding parameter)

  5. Multi dataset report – show statistics when running immuneML with the MultiDatasetBenchmarkTool

These types of reports are modeled by the following classes:

The existing reports can be found in the package reports. These can be specified in the YAML by specifying the name and optional parameters (see: How to specify an analysis with YAML).

Creating a custom report

Determine the type of report

First, it is important to determine what the type of the report is, as this defines which report class should be inherited.

If the report will be used to analyze a Dataset (such as a RepertoireDataset), either a DataReport or an EncodingReport should be used. The simplest report is the DataReport, which should typically be used when summarizing some qualities of a dataset. This dataset can be found in the report attribute dataset.

Use the EncodingReport when it is necessary to access the encoded_data attribute of a Dataset. The encoded_data attribute is an instance of a EncodedData class. This report should be used when the data representation first needs to be changed before running the report, either through an existing or a custom encoding (see: How to add a new encoding). For example, the Matches report represents a RepertoireDataset based on matches to a given reference dataset, and must first be encoded using a MatchedSequences, MatchedReceptors or MatchedRegex.

When the results of an experiment with a machine learning method should be analyzed, an MLReport or TrainMLModelReport should be used. These reports are a bit more advanced and require more understanding of the TrainMLModelInstruction. The MLReport should be used when plotting statistics or exporting information about one trained ML model. This report can be executed on any trained ML model, both in the assessment and selection loop of the TrainMLModel. An MLReport has the following attributes:

  1. train_dataset: a Dataset (e.g., RepertoireDataset) object containing the training data used for the given classifier

  2. test_dataset: similar to train_dataset, but containing the test data

  3. method: the MLMethod object containing trained classifiers for each of the labels.

  4. label: the label that the report is executed for (the same report may be executed several times when training classifiers for multiple labels), can be used to retrieve relevant information from the MLMethod object.

  5. hp_setting: the HPSetting object, containing all information about which preprocessing, encoding, and ML methods were used up to this point. This parameter can usually be ignored unless specific information from previous analysis steps is needed.

In contrast, TrainMLModelReport is used to compare several [optimal] ML models. This report has access to the attribute state: a TrainMLModelState object, containing information that has been collected through the execution of the TrainMLModelInstruction. This includes all datasets, trained models, labels, internal state objects for selection and assessment loops (nested cross-validation), optimal models, and more.

Finally, the MultiDatasetReport is used in rare cases when running immuneML with the MultiDatasetBenchmarkTool. This can be used when comparing the performance of classifiers over several datasets and accumulating the results. This report has the attribute instruction_states: a list of several TrainMLModelState objects.

Implementing the report

The new report should inherit the appropriate report type and be placed in the respective package (under reports, choose data_reports, encoding_reports, ml_reports, train_ml_model_reports, or multidataset_reports). The abstract method generate() must be implemented, which has the following responsibilities:

  • It should create the report results, for example, compute the data or create the plots that should be returned by the report.

  • It should write the report results to the folder given at the variable result_path.

  • It should return a ReportResult object, which contains lists of ReportOutput objects. These ReportOutput objects simply contain the path to a figure, table, text, or another type of result. One report can have multiple outputs, as long as they are all accessible through the ReportResult. This will be later used to format the summary of the results in the HTML output file.

  • When the main result of the report is a plot, it is good practice to also make the raw data available to the user, for example as a csv file.

The preferred method for plotting data is through plotly, as it creates interactive and rescalable plots in HTML format [recommended] that display nicely in the HTML output file. Alternatively, plots can also be in pdf, png, jpg and svg format.

Note

When plotting data with plotly, we recommend using the following color schemes for consistency: plotly.colors.sequential.Teal, plotly.colors.sequential.Viridis, or plotly.colors.diverging.Tealrose. Additionally, in the most of immuneML plots, ‘plotly_white’ theme is used for the background.

For the overview of color schemes, visit this link. For plotly themes, visit this link.

The second abstract method to be implemented is build_object(). This method can take in any custom parameters and should return an instance of the report object. The parameters of the method build_object() can be directly specified in the YAML specification, nested under the report type, for example:

MyNewReport:
  custom_parameter: “value”

Inside the build_object() method, you can check if the correct parameters are specified and raise an exception when the user input is incorrect (for example using the ParameterValidator utility class). Furthermore, it is possible to resolve more complex input parameters, such as loading reference sequences from an external input file, before passing them to the __init__() method of the report.

It is important to consider whether the method check_prerequisites() should be implemented. This method should return a boolean value describing whether the prerequisites are met, and print a warning message to the user when this condition is false. The report will only be generated when check_prerequisites() returns true. This method should not be used to raise exceptions. Instead, it is used to prevent exceptions from happening during execution, as this might cause lost results. Situations to consider are:

  • When implementing an EncodingReport, use this function to check that the data has been encoded and that the correct encoder has been used.

  • Similarly, when creating an MLReport or TrainMLModelReport, check that the appropriate ML methods have been used.

Note

When adding new features to immuneML, some utility classes are already available. For instance, to construct a path, you can use PathBuilder.build() function. If you need to validate some parameters when constructing an object in build_object() functions, for example, you can use ParameterValidator class. For the full list of such classes, see the util package.

Note

Please see the Report class for the detailed description of the methods to be implemented.

Test run of the report: specifying the new report in YAML

Custom reports may be defined in the YAML specification under the key ‘definitions’ the same way as any other reports. The easiest way to test run Data reports and Encoding reports is through the ExploratoryAnalysis instruction. They may also be specified in the TrainMLModel instruction in the selection and assessment loop under reports:data_splits and reports:encoding respectively.

ML model reports and Train ML model reports can only be run through the TrainMLModel instruction. ML reports can be specified inside both the selection and assessment loop under reports:models. Train ML model reports must be specified under reports.

Finally, Multi dataset reports can be specified under benchmark_reports when running the MultiDatasetBenchmarkTool.

The following specification shows the places where Data reports, Encoding reports , ML model reports, and Train ML model reports can be specified:

definitions:
  reports:
    my_data_report: MyNewDataReport # example data report without parameters
    my_encoding_report: # example encoding report with a parameter
      MyNewEncodingReport:
       parameter: value
    my_ml_report: MyNewMLReport # ml model report
    my_trainml_report: MyNewTrainMLModelReport # train ml model report

  datasets:
    d1:
      # if you do not have real data to test your report with, consider
      # using a randomly generated dataset, see the documentation:
      # “How to generate a random receptor or repertoire dataset”
      format: RandomRepertoireDataset
      params:
          labels: {disease: {True: 0.5, False: 0.5}}
          repertoire_count: 50
  encodings:
    e1: KmerFrequency
  ml_methods:
    m1: LogisticRegression

instructions:
  exploratory_instr: # Example of specifying reports in ExploratoryAnalysis
    type: ExploratoryAnalysis
    analyses:
      analysis_1: # Example analysis with data report
        dataset: d1
        report: my_data_report
      analysis_1: # Example analysis with encoding report
        dataset: d1
        encoding: e1
        report: my_encoding_report
        labels: # when running an encoding report, labels must be specified
            - disease

  trainmlmodel_instr: # Example of specifying reports in TrainMLModel instruction
    type: TrainMLModel
    settings:
      - encoding: e1
        ml_method: m1
    assessment: # running reports in the assessment (outer) loop
      reports:
        data: # execute before splitting to training/(validation+test)
          - my_data_report
        data_splits: # execute on training and (validation+test) sets
          - my_data_report
        encoding:
          - my_encoding_report
        models:
          - my_ml_report
    selection: # running reports in the selection (inner) loop
      reports:
        data: # execute before splitting to validation/test
          - my_data_report
        data_splits: # execute on validation and test sets
          - my_data_report
        encoding:
          - my_encoding_report
        models:
          - my_ml_report
    reports:
      - my_trainml_report
    labels:
      - disease

Adding documentation for the new report

After implementing the desired functionality, the documentation for the report should be added, so that the users of immuneML have sufficient information when deciding to use the report. It should be added to the docstring and consist of the following components:

  1. A short description of what the report is meant for.

  2. Optional extended description, including any references or specific cases that should bee considered.

  3. List of arguments the report takes as input. If the report does not take any arguments other than the ones provided by the immuneML in runtime depending on the report type (such as training and test dataset or trained method), there should be only a short statement that the report does not take input arguments.

  4. An example of how the report can be specified in YAML.

Documentation should be written in Sphinx reStructuredText formatting. Here is an example of a documentation for the DesignMatrixExporter report that has no input arguments which can be provided by the user in the YAML specification (the encoded dataset to be exported will be provided by immuneML at runtime):

Generates a histogram of the lengths of the sequences in a RepertoireDataset.

YAML specification:

.. indent with spaces
.. code-block:: yaml

    my_sld_report: NewSequenceLengthDistribution