Isovar

• Overview
• Python API
• Commandline
• Internal Design
• Other Isovar Commandline Tools
• Sequencing Recommendations

Overview

Isovar determines mutant protein subsequences around mutations from cancer RNAseq data.

Isovar works by:

1. collecting RNA reads which spanning the location of a variant,

2. filtering the RNA reads to those which support the mutation,

3. assembling mutant reads into longer coding sequences,

4. matching mutant coding sequences against reference annotated reading frames, and

5. translating coding sequences determined directly from RNA into mutant protein sequences.

The assembled coding sequences may incorporate proximal (germline and somatic) variants, along with any splicing alterations which occur due to modified splice signals.

Python API

In the example below, isovar.run_isovar returns a list of isovar.IsovarResult objects. Each of these objects corresponds to a single input variant and contains all of the information about the RNA evidence at that variant's location and any mutant protein sequences which were assembled for the variant.

from isovar import run_isovar

isovar_results = run_isovar(
    variants="cancer-mutations.vcf",
    alignment_file="tumor-rna.bam")
    
# this code traverses every variant and prints the number
# of RNA reads which support the alt allele for variants
# which had a successfully assembled/translated protein sequence
for isovar_result in isovar_results:
    # if any protein sequences were assembled from RNA
    # then the one preferred by the context/support policy can be
    # accessed from a property called `top_protein_sequence`.
    if isovar_result.top_protein_sequence is not None:
        # print number of distinct fragments supporting the
        # the variant allele for this mutation
        print(isovar_result.variant, isovar_result.num_alt_fragments)
    

A collection of IsovarResult objects can also be flattened into a Pandas DataFrame:

from isovar import run_isovar, isovar_results_to_dataframe

df =  isovar_results_to_dataframe(
        run_isovar(
            variants="cancer-mutations.vcf",
            alignment_file="tumor-rna.bam"))

RNA support versus vaccine context

Isovar 1.8.0 derives its context target from desired peptide size K: 2*K-1 (15mers → 29 aa; 25mers → 49 aa; 30mers → 59 aa). For a centered single-residue mutation this includes every mutation-containing Kmer. The default balanced policy maximizes actual mutation-overlapping windows among candidates retaining at least 85% of the best candidate's compatible read-name support, with an independent absolute floor of two read objects at each retained RNA base. Both thresholds are configurable. Actual context adapts to RNA support and coverage; protein boundaries and stops can also produce shorter output. This is not 85% of per-base depth.

This is a configurable selection tolerance, not a confidence estimate. Allele counts are unchanged, and no reference sequence is used to fill missing RNA. See context selection and configuration for support-first/context-first alternatives and the original tumor-RNA audit for real sequence comparisons. Vaxrank's explicit context-length request remains respected; its coordinated default change is tracked separately.

Python API options for collecting RNA reads

To change how Isovar collects and filters RNA reads you can create your own instance of the isovar.ReadCollector class and pass it to run_isovar.

from isovar import run_isovar, ReadCollector

# create a custom ReadCollector to change options for how RNA reads are processed
read_collector = ReadCollector(
    use_duplicate_reads=True,
    use_secondary_alignments=True, 
    use_soft_clipped_bases=True)

isovar_results = run_isovar(
    variants="cancer-mutations.vcf",
    alignment_file="tumor-rna.bam",
    read_collector=read_collector)

Since 1.17.0, read support counts sequenced segments, not their alternative SAM alignments (#264). Only complementary primary mates in the same read group are collapsed. Secondary placements remain available, but incompatible placements of one segment cannot extend a cDNA assembly or inflate support; conflicting allele calls from that segment are retained as uncertain (other_reads). Read-group-aware fragment counts are separate from the original string-valued *_read_names properties. Existing read constructors remain supported with optional provenance fields.

This can change counts, filtering and reconstructed context for multimapped reads; it does not establish independent molecules or perform UMI deduplication. Since 1.17.1, cross-variant phasing also uses read-group-scoped fragment IDs (#282). Matching names in different read groups cannot create an edge; paired segments count as one fragment. Public read-name helpers and phase-group names remain strings for display, not unique evidence IDs. Caller-created reads without provenance retain name-only phasing with other legacy reads; they are not equated with collected reads. Since 1.17.2, cross-variant phasing also requires compatible placements (#284): a shared fragment must have at least one compatible pair of observations, with no competing placements of the same segment. Complementary mates and variants on one spliced alignment still phase; incompatible alternatives cannot inflate thresholds or group support. Since 1.17.3, separately observed supplementary pieces can phase when reciprocal SA declarations establish the same chimeric path (#286). These are pieces of one sequenced segment, not paired mates or alternative placements. Strand and hard clipping are normalized to original-query coordinates; ambiguous path links or variant bases in overlapping pieces remain unphased. Minimap2's approximate SA CIGARs identify paths only: actual record CIGARs still determine alleles. A tag alone never creates supporting evidence, and ordinary cDNA assembly still requires one linear placement per segment. Groups remain connected pairwise evidence, not proof of one globally resolved haplotype.

Python API options for coding sequence assembly and translation

To change how Isovar assembles RNA reads into coding sequences, determines their reading frames, and groups translated amino acid sequences you can create your own instance of the isovar.ProteinSequenceCreator class and pass it to run_isovar.

As of Isovar 1.9.0, overlap assembly is enabled by default in both the Python API and command-line tools. To preserve the earlier Python API behavior, pass variant_sequence_assembly=False to ProteinSequenceCreator; on the command line, use --disable-variant-sequence-assembly.

As of Isovar 1.10.0, BAM-derived reads retain their aligned exon blocks and splice junctions during protein creation. Each read remains compatible with a set of transcripts; overlapping reads are assembled within shared compatible paths, and the resulting cDNA is translated only against those transcripts. Evidence that ends before an isoform-distinguishing junction remains ambiguous and can support every compatible branch without being counted more than once.

from isovar import run_isovar, ProteinSequenceCreator

# create a custom ProteinSequenceCreator to change options for how
# protein sequences are assembled from RNA reads
protein_sequence_creator = ProteinSequenceCreator(
    # number of amino acids we're aiming for, coding sequences
    # might still give us a shorter sequence due to an early stop 
    # codon or poor coverage
    protein_sequence_length=30,
    # minimum number of reads covering each base of the coding sequence
    min_variant_sequence_coverage=2,
    # how much of a reference transcript should a coding sequence match before
    # we use it to establish a reading frame
    min_transcript_prefix_length=20,
    # how many mismatches allowed between coding sequence (before the variant)
    # and transcript (before the variant location)
    max_transcript_mismatches=2,
    # also count mismatches after the variant location toward
    # max_transcript_mismatches
    count_mismatches_after_variant=False,
    # if more than one protein sequence can be assembled for a variant
    # then drop any beyond this number 
    max_protein_sequences_per_variant=1,
    # enabled by default; if set to False then coding sequence will be derived from
    # a single RNA read with the variant closest to its center
    variant_sequence_assembly=True,
    # how many nucleotides must two reads overlap before they are combined
    # into a single coding sequence
    min_assembly_overlap_size=30)

isovar_results = run_isovar(
    variants="cancer-mutations.vcf",
    alignment_file="tumor-rna.bam",
    protein_sequence_creator=protein_sequence_creator)

Python API for filtering results

You can filter a collection of IsovarResult objects by any of their numerical properties using the filter_thresholds option of the run_isovar function. The value expected for this argument is a dictionary whose keys have named like 'min_fraction_ref_reads' or 'max_num_alt_fragments' and whose values are numerical thresholds. Everything after the 'min_' or 'max_' at the start of a key is expected to be the name of a property of IsovarResult. Many of the commonly accessed properties regarding RNA read evidence follow the pattern:

{num|fraction}_{ref|alt|other}_{reads|fragments} 

For example, in the following code the results are filtered to have 10 or more alt reads supporting a variant and no more than 25% of the fragments supporting an allele other than the ref or alt.

from isovar import run_isovar

isovar_results = run_isovar(
    variants="cancer-mutations.vcf",
    alignment_file="tumor-rna.bam",
    filter_thresholds={"min_num_alt_reads": 10, "max_fraction_other_fragments": 0.25})    

for isovar_result in isovar_results:
    # print each variant and whether it passed both filters
    print(isovar_result.variant, isovar_result.passes_all_filters)

A variant which fails one or more filters is not excluded from the result collection but it has False values in its corresponding filter_values dictionary property and will have a False value for the passes_all_filters property.

If a result collection is flattened into a DataFrame then each filter is included as a column.

It's also possible to filter on boolean properties (without numerical thresholds) by passing filter_flags to run_isovar. These boolean properties can be further negated by prepending 'not_' to the property name, so that both 'protein_sequence_matches_predicted_mutation_effect' and 'not_protein_sequence_matches_predicted_mutation_effect' are valid names for filter_flags.

Structural variants

The ordinary variant-to-protein pipeline accepts literal nucleotide alleles, including sequence-resolved indels. Symbolic structural variants (<DEL>, <DUP>, etc.), breakends, and varcode.StructuralVariant objects are rejected explicitly; their placeholder bases must not be interpreted as small variants. For supplied fusion RNA, isovar fusion --input fusion.json --output result.json validates junction evidence and annotated coding frames, retaining unresolved or ambiguous outcomes. See the fusion input/output contract.

Commandline

Basic example:

$ isovar run \
    --vcf somatic-variants.vcf  \
    --bam rnaseq.bam \
    --protein-sequence-length 30 \
    --output isovar-results.csv

isovar --help lists the subcommands. Each subcommand's --help lists its options and current defaults:

isovar --help
isovar run --help
isovar reference-contexts --help

For example, use only primary alignments, include soft-clipped bases, and require at least three read objects at every retained cDNA base:

isovar run --vcf somatic-variants.vcf --bam rnaseq.bam \
    --drop-secondary-alignments --use-soft-clipped-bases \
    --min-variant-sequence-coverage 3 --num-rna-decompression-threads 4 \
    --output isovar-results.csv

To export every candidate protein, use --max-protein-sequences-per-variant 0. The default keeps the top candidate for each variant. The isovar variant-sequences command accepts --variant-sequence-length to set its preferred cDNA length.

Existing scripts remain supported: isovar --vcf ... --bam ... still runs the main pipeline, and hyphenated commands such as isovar-protein-sequences and isovar-plot remain aliases. Both spellings use the same handlers and defaults. You can also invoke the CLI as python -m isovar.

Shared CLI and Python defaults

Defaults are defined in default_parameters.py. As of 1.11.0, CLI-created collectors and ReadCollector() merge overlapping mates, matching run_isovar(). Merged reads retain the number of contributing alignments in source_read_count. To keep mates separate, pass --no-merge-overlapping-fragments or ReadCollector(merge_overlapping_fragments=False). Mates with conflicting alignment paths remain separate. For matching paths, the existing consensus rule resolves disagreeing bases by quality; this can change allele support as well as assembled sequences.

The CLI now applies the complete default filter set used by run_isovar(), including read-level allele fractions and limits on other-allele support. This adds filter columns to CLI output and can change passes_all_filters. Explicit Python filter_thresholds dictionaries still replace the defaults; to override selected defaults, copy DEFAULT_FILTER_THRESHOLDS and update it. The existing imports from isovar.main remain supported.

--reference-context-size belongs only to isovar reference-contexts, where it must be positive. Protein-producing commands now reject this previously ignored option. They derive reference context size from the requested cDNA length and minimum transcript prefix. Automatic protein length and Vaxrank's explicit peptide/context settings remain supported.

Internal Design

The inputs to Isovar are one or more somatic variant call (VCF) files, along with a BAM file containing aligned tumor RNA reads. The following objects are used to aggregate information within Isovar:

• LocusRead: Isovar examines each variant locus and extracts reads overlapping that locus, represented by LocusRead. The LocusRead representation allows filtering based on quality and alignment criteria (e.g. MAPQ > 0) which are thrown away in later stages of Isovar.

• AlleleRead: Once LocusRead objects have been filtered, they are converted into a simplified representation called AlleleRead. Each AlleleRead contains only the cDNA sequences before, at, and after the variant locus.

• ReadEvidence: The set of AlleleRead objects overlapping a mutation's location may support many different distinct allele. The ReadEvidence type represents the grouping of these reads into ref, alt and other AlleleRead sets, where ref reads agree with the reference sequence, alt reads agree with the given mutation, and other reads contain all non-ref/non-alt alleles. The alt reads will be used later to determine a mutant coding sequence, but the ref and other groups are also kept in case they are useful for filtering.

• VariantSequence: Overlapping AlleleReads containing the same mutation are assembled into a longer sequence. The VariantSequence object represents this candidate coding sequence, as well as all the AlleleRead objects which were used to create it.

• ReferenceContext: To determine the reading frame in which to translate a VariantSequence, Isovar looks at all Ensembl annotated transcripts overlapping the locus and collapses them into one or more ReferenceContext object. Each ReferenceContext represents the cDNA sequence upstream of the variant locus and in which of the {0, +1, +2} reading frames it is translated.

• Translation: Use the reading frame from a ReferenceContext to translate a VariantSequence into a protein fragment, represented by Translation.

• ProteinSequence: Multiple distinct variant sequences and reference contexts can generate the same translations, so we aggregate those equivalent Translation objects into a ProteinSequence.

• IsovarResult: Since a single variant locus might have reads which assemble into multiple incompatible coding sequences, an IsovarResult represents a variant and one or more ProteinSequence objects which are associated with it. Protein sequences are ranked by the configured context/support preference and the top sequence is made easy to access. Allele-support properties such as num_alt_fragments and fraction_ref_reads remain separate from the selected protein's compatible support.

Other Isovar Commandline Tools

isovar plot renders white-background protein, coverage, read-overlap and local transcript figures as individual SVGs and 600-dpi PNGs, plus an overview, in UTC date/time-stamped directories. Install isovar[plot], then see the plotting commands and reproducible osteosarc assembly examples.

isovar protein-sequences --vcf variants.vcf --bam rna.bam

Candidate protein sequences from RNA reads; keeps the top sequence per variant unless --max-protein-sequences-per-variant 0 is supplied.

isovar allele-counts --vcf variants.vcf --bam rna.bam

Counts of reads and fragments supporting the ref, alt, and other alleles at all given variant locations.

isovar allele-reads --vcf variants.vcf --bam rna.bam

Sequences of all reads overlapping any of the given variants.

isovar translations --vcf variants.vcf --bam rna.bam

All possible translations of any assembled cDNA sequence containing any of the given variants in the reference frame of any matching transcript.

isovar reference-contexts --vcf variants.vcf

Shows all candidate reference contexts (sequence and reading frame) before each variant, derived from overlapping reference coding transcripts.

isovar variant-reads --vcf variants.vcf --bam rna.bam

Like the isovar allele-reads command but limited only to reads which support the alt allele.

isovar variant-sequences --vcf variants.vcf --bam rna.bam

Shows all assembled cDNA coding sequences supporting any of the given variants.

Sequencing Recommendations

Isovar works best with high quality / high coverage mRNA sequence data. This means that you will get best results from >100M paired-end reads sequenced on an Illumina HiSeq from a library enriched with poly-A capture. The number of reads varies depending on degree of RNA degradation and tumor purity. The read length will determine the longest protein sequence you can recover, since Isovar's cDNA assembly only considers reads that overlap a variant. With 100bp reads you will be able to assemble at most 199bp of sequence around a somatic single nucleotide variant, and consequently only be to determine 66 amino acids from the protein sequence. If you disable the cDNA assembly algorithm then a 100bp read will only be able to determine 33 amino acids.

Note on long-read and error-prone data: Isovar's cDNA assembly algorithm requires exact sequence matches when detecting overlaps between reads. This is well-suited for Illumina short reads (~0.1% error rate) but will produce fragmented or incomplete assemblies with long-read technologies (PacBio, Oxford Nanopore) that have higher indel error rates. The coverage-trimming step also assumes that read coverage decreases monotonically away from the variant locus, which may not hold for reads spanning splice junctions.