FilterByOrientationBias is not aggressive enough when two artifact modes are specified and one has a suppression factor of 0.0

[LeeTL1220]

In this case, the FDR threshold is not honored. The explanation of this is complex, but essentially has to do with the Benjamini-Hochberg procedure not playing well with suppression factor when extended to more than one artifact mode.

The definition of FilterByOrientationBias will have to be changes from "guaranteeing less than 1% FDR over all mutations" to "guaranteeing less than 1% FDR in each specified artifact mode". This could make the filter more aggressive, so we may have to adjust the FDR threshold.

• code fix
• doc updates

[LeeTL1220]

I have a real example that is close to the below fake example:

Say you have:

• 1000 OxoG mode variants and suppression of 1.0 (lots of artifacts)
• 40 FFPE mode variants with suppression of 0.0 (no artifacts).
• 100 variants that are not in any artifact mode.
• FDR Threshold 0.01

In this fake example (where I don't show the p-values, which is an important piece of this) up cutting:

• none of the 140 mutations (FFPE + non-artifact mode)
• about 980 of the OxoG mode artifacts.

Unfortunately, this leaves about 20 OxoG artifact modes and a FDR that is much higher than 0.01. This is because the FFPE p-values are considered, which saves a bunch of the OxoG variants from filtering, and then the suppression factor is applied.