Replace stampy with minimap2

[bricoletc]

@iqbal-lab could you give me access to the cortex repo? so I could in future push code to a minimap2 branch for eg.
In the meantime this is for reviewing my changes.

Code

• Remove stampy command line args in run_calls.pl and
  process_calls.pl, replacing with a single minimap2_bin argument
• instead of stampy index, use ref fasta (passed as --ref_fasta)
  to place variants in process_calls.pl (minimap2 can index ref on the
  fly)
  Previously, --ref_fasta was used to check for misplaced variants. run_calls.pl always uses the option. Now, this always occurs, as --ref_fasta is used to place variants.

Checks

I have validated that I get the same VCF when running py_cortex_api (it wraps cortex's independent workflow) with stampy and with minimap2, on cortex's demo files example1

TODOs

• A systematic check on several datasets- notably, I observed minimap2 MAPQ seems lower than stampy for 50bp reads
• Update cortex manual
• Maybe this code can be removed, as minimap2 could probably map >1000bp sequence even in short-read preset? :

  cortex/scripts/analyse_variants/process_calls.pl

  Lines 2682 to 2691 in 6d8e2f9

  	## if the 5p flank is >1000bp, then stampy fails, so we cut the 5p flank and only take the last 1000bp.
  	if ( exists $href_var_name_to_cut_flank->{$name} )
  	{
  	##the flank5p passedin is directly taken from the callfile
  	$flank5p = substr( $flank5p, -1000 );
  	if ( length($flank5p) != 1000 )
  	{
  	die("perl issue with $flank5p");
  	}
  	}

[bricoletc]

Logging here a validation evaluation of using minimap2. Evaluation uses varifier on one staph dataset from martin (with truth genome assembled using pilon), :

I found that the 'good' minimap2 settings (-k9 -w4) give following recall/precision:
  * require mapq > 40: "Precision_edit_dist": 0.99996956, "Recall_edit_dist": 0.50726812
  * require mapq > 0:  "Precision_edit_dist": 0.99998191, "Recall_edit_dist": 0.83881474,
  * previous cortex, using stampy and mapq > 40, gets "Precision_edit_dist": 0.98863584, "Recall_edit_dist": 0.83029847,
on this dataset, minimap2 with mapq>0 does a little better in both metrics.

[bricoletc]

Logging some more validation work using 14 ilmn Plasmodium falciparum samples with matches pacb assemblies (in previous comment, ran on a Staphylococcus aureus dataset):

stampy: recall: 0.4490 precision: 0.9118
minimap2: recall: 0.4443 precision: 0.9258

These are the average of metrics computed using varifier. THis confirms the change works fine.

[iqbal-lab]

hurrah