Streamline predicate/analysis workflow

[ielis]

Fixes #87 , #92

Depends on #94

[ielis]

@lnrekerle

I'm proposing a revamp to the CohortAnalyzer.

The CohortAnalyzer is an abstraction - a promise what CohortAnalyzer can do for the user. To get CohortAnalyzer we use a similar pattern to configuring PhenopacketPatientCreator. There is a config method that will give you CohortAnalyzer:

from genophenocorr.analysis import configure_cohort_analysis

analysis = configure_cohort_analysis(cohort, hpo)

You'll get an analysis with default options. If you want to tweak the options, build the CohortAnalysisConfiguration:

from genophenocorr.analysis import CohortAnalysisConfiguration

configuration = CohortAnalysisConfiguration.builder()
  .include_sv(True)
  .pval_correction('fdr_bh')
  .build()

analysis = configure_cohort_analysis(cohort, hpo, configuration)

Then we run the analysis, e.g. to compare MISSENSE vs others:

from genophenocorr.model import VariantEffect
from genophenocorr.analysis.predicate import BooleanPredicate

results = analysis.compare_by_variant_effect(VariantEffect.MISSENSE_VARIANT, tx_id='NM_1234.5')
result_df = results.summarize(hpo, BooleanPredicate.YES)
result_df.head()

We get results, a container with a lot of data. We call summarize to prepare a data frame with phenotypes vs. genotypes, ordered by corrected p values.

Note that we provide BooleanPredicate.YES to show genotype-phenotype correlation for present HPO terms, not for not-present (we would use BooleanPredicate.NO to show those).

This is what the PR adds. Thanks to the changes, we have a general framework for applying genotype and phenotype predicates and showing the results.

Please check out the code, try it out and we can discuss in greater detail the next time.

[ielis]

Now, with the develop merged into the PR branch, we should be OK to move forward with this PR if the code looks good.