Merge branch 'release/v5.1.1'

ACEnglish · Feb 5, 2025 · 14fa7b6 · 14fa7b6
2 parents 7a45d98 + 61248ee
commit 14fa7b6
Show file tree

Hide file tree

Showing 153 changed files with 9,103 additions and 7,463 deletions.
diff --git a/docs/Updates.md b/docs/Updates.md
@@ -1,6 +1,37 @@
-# Truvari 5.0
+# Truvari 5.1.1
 *in progress*
 
+* `bench`
+  *  new automatic hook into the refine step via `truvari bench --refine`
+* `refine`
+  * completely reworked UI in favor of easier whole-genome SV refinement. See wiki for details
+  * Now writes a consolidated `refine.base.vcf.gz` and `refine.comp.vcf.gz` for easier tracking of variants' final states.
+  * Default behavior count original variant representations instead of the `phab` variant representations 
+* `collapse`
+  * Add `--dup-to-ins`
+  * Fixed bug where regions with >100 variants would sometimes not have all variants compared
+  * `--chain` functionality now capped to do only 1 transitive match, preventing uncontrolled over-merging
+* `ga4gh`
+  * New/renamed parameters as part of general improvement work
+  * Output suffixes are now `.base.vcf.gz` and `.comp.vcf.gz` for consistency.
+* `stratify` 
+  * 1--complement` now outputs a single line of total variant counts outside of the regions instead of arbitrarily assigning variants to their nearest region
+
+* misc
+  * Fix BND bugs
+    * `pysam.VariantFile.allele_variant_types` falsely identified some BNDs as INDELs, causing incorrect filtering by Truvari
+    * SVs Decomposed to BNDs strandedness flipped to be more representative of original SV 
+  * unroll seqsim checks all directions
+  * Match sorting breaks seq/size ties with start/end distance
+  * Long SV roll limit speeds - ≥500bp, rolling is turned off
+  * `truvari.VariantRecord.within` edge case fix
+
+
+
+
+# Truvari 5.0
+*January 9, 2025*
+
 * Reference context sequence comparison is now deprecated and sequence similarity calculation improved by also checking lexicographically minimum rotation's similarity. [details](https://github.com/ACEnglish/truvari/wiki/bench#comparing-sequences-of-variants)
 * Symbolic variants (`<DEL>`, `<INV>`, `<DUP>`) can now be resolved for sequence comparison when a `--reference` is provided. The function for resolving the sequences is largely similar to [this discussion](https://github.com/ACEnglish/truvari/discussions/216)
 * Symbolic variants can now match to resolved variants, even with `--pctseq 0`, with or without the new sequence resolving procedure.

diff --git a/docs/bench.md b/docs/bench.md
@@ -178,17 +178,33 @@ Truvari can replace the symbolic alt of resolved SVs in the output VCF with the
 
 BND Comparison
 ==============
-Breakend (BND) variants are compared by checking a few conditions using a single threshold of `--bnddist` which holds the maximum distance around a breakpoint position to search for a match. Similar to the `--refdist` parameter, truvari looks for overlaps between the `dist` 'buffered' boundaries (e.g. `overlaps( POS_base - dist, POS_base + dist, POS_comp - dist, POS_comp + dist)` 
+Breakend (BND) variants are compared by checking a few conditions using a single threshold of `--bnddist` which holds the maximum distance around a breakpoint position to search for a match. Similar to the `--refdist` parameter, truvari looks for overlaps between the `dist` 'buffered' boundaries (e.g. `overlaps( POS_base - dist, POS_base + dist, POS_comp - dist, POS_comp + dist)` Additionally, if the CIPOS and and CIEND info tags are available in the entry, the e.g. POS is further buffered by `-abs(CIPOS[0])` and `+(abs(CIPOS[1])`.
 
 The baseline and comparison BNDs' POS and their joined position must both be within `--bnddist` to be a match candidate (i.e. no partial matches). Furthermore, the direction and strand of the two BNDs must match, for example `t[p[` (piece extending to the right of p is joined after t) only matches with `t[p[` and won't match to `[p[t` (reverse comp piece extending right of p is joined before t). 
 
 BND's are annotated in the truvari output with fields:  StartDistance (baseline minus comparison POS); EndDistance (baseline minus comparison join position);  TruScore which describes the percent of the allowed distance needed to find this match (`(1 - ((abs(StartDistance) + abs(EndDistance)) / 2) / (bnddist*2)) * 100`). For example, two BNDs 20bp apart with bnddist of 100 makes a score of 90.
 
-Another complication for matching BNDs is that they may represent an event which could be 'resolved' in another VCFs. For example, a tandem duplication between `start-end` could be represented as two BNDs of `start to N[{chrom}:{end}[` and `end to ]{self.chrom}:{start}]N`. Therefore, truvari also attempts to compare symbolic alt SVs (ALT = `<DEL>`, `<INV>`, `<DUP>`) to a BND by decomposing the symbolic alt into its breakpoints. These decomposed BNDs are then each checked against a comparison BND and the highest TruScore match kept. 
+BND comparison can be turned off by setting `--bnddist -1`. Single-end BNDs (e.g. ALT=`TTT.`) are still ignored.
 
-Note that DUPs are always decomposed to DUP:TANDEM breakpoints. Note that with `--pick single`, a decomposed SV will only match to one BND, so `--pick multi` is recommended to ensure all BNDs will match to a single decomposed SV.
+Cross-Representation Matching
+=============================
 
-BND comparison can be turned off by setting `--bnddist -1`. Symbolic ALT decomposition can be turned off with `--no-decompose`. Single-end BNDs (e.g. ALT=`TTT.`) are still ignored.
+Truvari considers there to be three possible representation styles of SVs. 
+
+1. Resolved: SVs with the full REF and ALT sequences, most frequently representing INS and DEL. 
+2. Symbolic: SVs without the REF or ALT sequences having an ALT of e.g. `<DEL>, <DUP>`, etc. 
+3. BNDs: SV breakends represented with the e.g. `t[p[` ALT field.
+
+Comparing SVs across these representation styles have the following caveats:
+
+1. When comparing Resolved and Symbolic SVs, sequence similarity is turned off for thresholding matches. If a user provides a `--reference`, symbolic SVs shorter than the `--max-resolve` parameter (default 25kbp) can be turned into Resolved SVs [details in API docs](https://truvari.readthedocs.io/en/latest/truvari.package.html#truvari.VariantRecord.resolve) and therefore the sequence similarity thresholds are still enforced.
+2. When a BND is compared to a with Resolved or Symbolic SV, the SV is 'decomposed' into a set of BNDs and each is compared with the original BND. If any of the decomposed BNDs matches to the original BND, the Resolved/Symbolic SV and BND are considered matching. Details of SV decomposition are [in the API docs](https://truvari.readthedocs.io/en/latest/truvari.package.html#truvari.VariantRecord.decompose)
+
+Note that only Deletions (symbolic or resolved), INV (symbolic or resolved), and symbolic DUPs can be decomposed into BNDs. DUPs are always decomposed into DUP:TANDEM breakends. 
+
+Because SVs decompose into multiple BNDs (2 for DEL/DUP, 4 for INV), and because `--pick single` is the default, a decomposed SV will only match to one BND and the BNDs 'mate' will be a FN. To enable all BNDs to match to a decomposed SV, specify `--pick multi`.
+
+SV decomposition into BNDs can be turned off with `--no-decompose`. 
 
 Controlling the number of matches
 =================================

diff --git a/docs/collapse.md b/docs/collapse.md
@@ -8,6 +8,7 @@ To start, we merge multiple VCFs (each with their own sample) and ensure there a
 ```bash
 bcftools merge -m none one.vcf.gz two.vcf.gz | bgzip > merge.vcf.gz
 ```
+WARNING! If you have symbolic variants, see [the below section](https://github.com/ACEnglish/truvari/wiki/collapse#symbolic-variants) on using bcftools.
 
 This will `paste` SAMPLE information between vcfs when calls have the exact same chrom, pos, ref, and alt.
 For example, consider two vcfs:
@@ -40,6 +41,26 @@ For example, if we collapsed our example merge.vcf by matching any calls within
     >> truvari_collapsed.vcf
     chr1 7 ... GT ./. 0/1    
 
+Symbolic Variants
+=================
+bcftools may not handle symbolic variants correctly since it doesn't consider their END position. To correct for this, ensure that every input variant has a unique ID and use `bcftools merge -m id`. For example:
+```
+# A.vcf
+chr1	147022730	SV1	N	<DEL>	.	PASS	SVLEN=-570334;END=147593064
+# B.vcf
+chr1	147022730	SV2	N	<DEL>	.	PASS	SVLEN=-990414;END=148013144
+
+# bcftools merge -m none A.vcf B.vcf
+# Premature collapse
+chr1	147022730	SV1;SV2	N	<DEL>	.	PASS	SVLEN=-570334;END=147593064
+
+# bcftools merge -m id A.vcf B.vcf
+chr1	147022730	SV1	N	<DEL>	.	PASS	SVLEN=-570334;END=147593064
+chr1	147022730	SV2	N	<DEL>	.	PASS	SVLEN=-990414;END=148013144
+```
+
+This bug has been replicated with bcftools 1.18 and 1.21.
+
 --choose behavior
 =================
 When collapsing, the default `--choose` behavior is to take the `first` variant by position from a cluster to
@@ -89,18 +110,22 @@ will become:
 Normally, every variant in a set of variants that are collapsed together matches every other variant in the set. However, when using `--chain` mode, we allow 'transitive matching'. This means that all variants match to only at least one other variant in the set. In situations where a 'middle' variant has two matches that don't match each other, without `--chain` the locus will produce two variants whereas using `--chain` will produce one.
 For example, if we have
 
-    chr1 5 ..
+    chr1 1 ..
+    chr1 4 ..
     chr1 7 ..
-    chr1 9 ..
+    chr1 10 ..
 
-When we collapse anything within 2bp of each other, without `--chain`, we output:
+We take the `chr1 1` variant and find all its matches. When we collapse anything within 5bp of each other, without `--chain`, we output:
 
-    chr1 5 ..
-    chr1 9 ..
+    chr1 1 ..
+    chr1 7 ..
+
+With `--chain`, we would allow one level of transitive matching. This means that after finding the `chr1 1 -> chr1 4` match, we check `chr1 4` against all the remaining variants and would output
 
-With `--chain`, we would collapse `chr1 9` as well, producing
+    chr1 1 ..
+    chr1 10 ..
 
-    chr1 5 ..
+Note that this leaves `chr1 10` because we don't do multiple levels of transitive matching, meaning we never compare `chr1 7` to `chr1 10`. This is preferred because otherwise variants which have a continuous range of similarity could all be collapsed into a single variant. e.g., if the position in this example were sizes and, we wouldn't want the 1bp variant being a kept representation for all the variants.
 
 Annotations
 ===========
@@ -111,55 +136,4 @@ The output file has only two annotations added to the `INFO`.
 - `NumCollapsed` - Number of variants collapsed into this variant
 - `NumConsolidated` - Number of samples' genotypes consolidated into this call's genotypes
 
-The collapsed file has all of the annotations added by [[bench|bench#definition-of-annotations-added-to-tp-vcfs]]. Note that `MatchId` is tied to the output file's `CollapseId`. See [MatchIds](https://github.com/spiralgenetics/truvari/wiki/MatchIds) for details.
-
-```
-usage: collapse [-h] -i INPUT [-o OUTPUT] [-c COLLAPSED_OUTPUT] [-f REFERENCE] [-k {first,maxqual,common}] [--debug]
-                [-r REFDIST] [-p PCTSIM] [-B MINHAPLEN] [-P PCTSIZE] [-O PCTOVL] [-t] [--use-lev] [--hap] [--chain]
-                [--no-consolidate] [--null-consolidate NULL_CONSOLIDATE] [-s SIZEMIN] [-S SIZEMAX] [--passonly]
-
-Structural variant collapser
-
-Will collapse all variants within sizemin/max that match over thresholds
-
-options:
-  -h, --help            show this help message and exit
-  -i INPUT, --input INPUT
-                        Comparison set of calls
-  -o OUTPUT, --output OUTPUT
-                        Output vcf (stdout)
-  -c COLLAPSED_OUTPUT, --collapsed-output COLLAPSED_OUTPUT
-                        Where collapsed variants are written (collapsed.vcf)
-  -f REFERENCE, --reference REFERENCE
-                        Indexed fasta used to call variants
-  -k {first,maxqual,common}, --keep {first,maxqual,common}
-                        When collapsing calls, which one to keep (first)
-  --debug               Verbose logging
-  --hap                 Collapsing a single individual's haplotype resolved calls (False)
-  --chain               Chain comparisons to extend possible collapsing (False)
-  --no-consolidate      Skip consolidation of sample genotype fields (True)
-  --null-consolidate NULL_CONSOLIDATE
-                        Comma separated list of FORMAT fields to consolidate into the kept entry by taking the first non-null
-                        from all neighbors (None)
-
-Comparison Threshold Arguments:
-  -r REFDIST, --refdist REFDIST
-                        Max reference location distance (500)
-  -p PCTSIM, --pctsim PCTSIM
-                        Min percent allele sequence similarity. Set to 0 to ignore. (0.95)
-  -B MINHAPLEN, --minhaplen MINHAPLEN
-                        Minimum haplotype sequence length to create (50)
-  -P PCTSIZE, --pctsize PCTSIZE
-                        Min pct allele size similarity (minvarsize/maxvarsize) (0.95)
-  -O PCTOVL, --pctovl PCTOVL
-                        Min pct reciprocal overlap (0.0) for DEL events
-  -t, --typeignore      Variant types don't need to match to compare (False)
-  --use-lev             Use the Levenshtein distance ratio instead of edlib editDistance ratio (False)
-
-Filtering Arguments:
-  -s SIZEMIN, --sizemin SIZEMIN
-                        Minimum variant size to consider for comparison (50)
-  -S SIZEMAX, --sizemax SIZEMAX
-                        Maximum variant size to consider for comparison (50000)
-  --passonly            Only consider calls with FILTER == PASS
-```
+The collapsed file has all of the annotations added by [[bench|bench#definition-of-annotations-added-to-tp-vcfs]]. Note that `MatchId` is tied to the output file's `CollapseId`. See [MatchIds](https://github.com/spiralgenetics/truvari/wiki/MatchIds) for details.