Finish `AutoPP3BP4`

[gromdimon]

Is your feature request related to a problem? Please describe. After implementaino of #77 we have AutoPP3BP4 class. Now we need to properly implement all the methods and test them

Describe the solution you'd like

• Implement _get_best_pathogenic_score _get_best_benign_score and _predict_spliceai methods in AutoPP3BP4 class.
• Add unit tests for all this methods in tests/criteria/test_auto_pp4_bp4.py
• Add integreation tests to the tests/assets/e2e_vbariants/other_criteria.csv
• Add docstrings to all the methods in AutoPP3BP4

Describe alternatives you've considered N/A

Additional context Information about PP3 and BP4

PP3 (in silico predictions)

Original Definition

Multiple lines of computational evidence support a deleterious effect on the gene or gene product (conservation, evolutionary, splicing impact, etc).

Caveats:

- As many in silico algorithms use the same or very similar input for their predictions, each algorithm should not be counted as an independent criterion.
- PP3 can be used only once in any evaluation of a variant.

-- Richards et al. (2015); Table 4

Preconditions / Precomputations

• If the criterion PVS1 was triggered then this criterion is skipped.
• If the variant is on chrMT then it is skipped, as we don't have calibration for chrMT yet.
• If the variant is not found in dbNSFP or CADD precomputed scores then it is skipped as we don't have calibration for chrMT yet.

Implemented Criterion

An initial prediction is fist done using the general purpose pathogenicity predictors.

• If we have a score from the following, then the prediction is used (in descending order of priority):
  • REVEL, MutPred2, CADD, BayesDel, VEST4, ..., PhyloP
  • we will use the modifiers from :footcite:t:pejaver:2022
• If predictions are missing then then PhyloP of the position of the variant is used as a fallback.

Then, for splicing the following is done.

• If a SpliceAI prediction is performed then it is interpreted according to :footcite:t:walker:2023.

The highest-scoring variant is used for the final prediction.

User Report

• The scores and predictions from the predictors.

Caveats

• As described in :ref:acmg_seqvars_criteria-patho-predictions, we are currently limited to the precomputed threshold from the literature. This hinders us in adopting AlphaMissense effectively, for example.
• We need to compute accuracy to rank the implemented methods.
• We need our own calibration for chrMT.

Notes

• This criterion is similar to :ref:acmg_seqvars_criteria-bp4

BP4

BP4 (in silico predictions)

.. note::

- we have not implemented MitoTip or MitImpact yet
- we are lacking phylop scores yet
- we don't have live CADD scores yet

Original Definition

Multiple lines of computational evidence suggest no impact on gene or gene product (conservation, evolutionary, splicing impact, etc).

Caveat: As many in silico algorithms use the same or very similar input for their predictions, each algorithm cannot be counted as an independent criterion.
BP4 can be used only once in any evaluation of a variant.

-- Richards et al. (2015); Table 4

Preconditions / Precomputations

• If the criterion BA1 triggered then this criterion is skipped.
• If the variant is on chrMT then it is skipped, as we don't have calibration for chrMT yet.
• If the variant is not found in dbNSFP or CADD precomputed scores then it is skipped as we don't have calibration for chrMT yet.

Implemented Criterion

See :ref:acmg_seqvars_criteria-pp3 for details.

User Report

See :ref:acmg_seqvars_criteria-pp3 for details.

Literature

See :ref:acmg_seqvars_criteria-pp3 for details.

Caveats

See :ref:acmg_seqvars_criteria-pp3 for details.

Notes

• This criterion is similar to :ref:acmg_seqvars_criteria-pp3

Intervar

Intervar

PP3 and BP4 by Automated Scoring When multiple pieces of computational evidence support a deleterious effect on the gene or gene product (conservation, evolutionary, splicing impact, etc.), then the supporting pathogenic evidence of PP3 will be assigned as 1. In comparison, when multiple pieces of computational evidence suggest no impact on the gene or gene product, then supporting benign evidence of BP4 will be assigned as 1. All sets of in silico results must agree when PP3 or BP4 is assigned. These multiple pieces of computational evidence can be provided by ANNOVAR from the “dbnsfp30a” database, where the MetaSVM score16 is used for deleteriousness prediction and GERP++ is used for evolutionary conservation. The splicing impacts can be inferred by ANNOVAR from the “dbscsnv11” database. For the evidence of PP3 and BP4, we set the cutoff to 0.0 for MetaSVM scores (greater scores indicate more likely deleterious effects), 2.0 for GERP++_RS (smaller scores indicate less conservation), and 0.6 for adaptive boosting (ADA) and random forest (RF) scores of dbscSNV as splicing impact (larger scores indicate more likely splice altering).

[gromdimon]

Image with scores: https://www.ncbi.nlm.nih.gov/pmc/articles/instance/9748256/bin/gr4.jpg