1. Masked and inactivating mutations
2. Mutation data in the output files
   1. Mutation report (inactivating_mutations.tsv)
   2. Mutation plots
   3. Decorator browser track
3. Frame-shifting insertions and deletions
   1. Frame shift compensation
4. Premature stop codons (nonsense mutations)
5. Splice site mutations
6. Big insertions and deletions
7. Missing start codon
8. Missing stop codon
9. Missing and deleted exons
   1. Missing exon
   2. Deleted exon
10. Reference-absent introns

During the alignment step, TOGA2 checks the annotated projection sequences for potential gene-inactivating mutations. The number, location, and profile of the annotated mutations impact the projection loss status classification, which, in its turn, affects the transcript- and gene loss analysis, query gene inference, and orthology relationship resolution at the latter TOGA2 steps.

While all the mutation events recorded during the alignment step are colloquially referred to as inactivating, their actual impact on gene integrity may vary, also depending on their location in the coding sequence: a frameshift or premature stop codon in the middle of the gene severely truncates the protein, whereas a frameshift or stop codon situated close to the end of the coding sequence just affects the C-terminus.

To reflect the difference in mutation effect on gene function, TOGA2 uses the concept of “masked mutations”, which are considered to have a less pronounced or nonexistent effect on gene integrity, and can be completely ignored or treated more leniently during gene loss classification. Mutations that are not masked are considered as potentially gene- inactivating mutations.

Common reasons for mutation masking include:

• Mutations located in the first (N-terminal) 10% of the reference CDS, provided an alternative start codon is present. In this case, TOGA2 assumes that translation starts from the downstream start codon, resulting in a functional protein, albeit with a truncated N-terminus. The alternative start codon must lie in the same first 10% of the coding sequence and be in the same reading frame as the original reference start codon.
  • The requirement for the presence of an alternative start codon for N-terminal mutations can be lifted by setting the --mask_n_terminal_mutation/-m10m flag. In this case, all mutations in the N-terminal 10% of the CDS are masked.
• Mutations located in the last (C-terminal) 10% of the reference CDS. Masking N- and C-terminal mutations is based on the observation that frameshifts and premature stop codons in conserved human-mouse orthologs are heavily biased toward the first and last 10% of the CDS.
• Mutations located in exons classified as Missing or Deleted in the query. Note that the exon absence itself is still considered an inactivating mutation event (see below).

The following mutations or events are always masked and ignored when inferring the projection loss status:

• Non-canonical splice site dinucleotides (e.g., differing from GT/GC consensus for the donor site and the AG consensus dinucleotide for the acceptor site) if the respective counterpart in the reference is a U12 or a non-canonical U2 intron;
• Precise intron deletions in the query;
• Reference-absent introns annotated using SpliceAI predictions;
• Splice site shifts inferred from SpliceAI probabilities;
• Big (≥50 bases) insertions and deletions, including those introduced by SpliceAI-supported splice site correction;
• Frame-shifting indels introduced during the SpliceAI-supported splice site correction, and frame-shifts in the original CESAR2 alignment compensated by these indels.

Information on mutations and events found in the annotated projections is stored in the inactivating_mutations.tsv file, a tab-delimited table file with the following columns:

• projection: projection name;
• exon: affected exon number with respect to the reference exon-intron structure;
• triplet: number of the triplet (three consecutive symbols in the query) where the mutation occurred. If a mutation affects multiple triplets, the value contains the first and the last triplet numbers separated by underscore.
• ref_codon: reference codon number corresponding to the affected query codon. If a mutation affects multiple reference codons, the value contains the first and the last codon numbers separated by underscore.
• chrom: name of the query chromosome (contig, scaffold, etc.) where the mutation occurred;
• start: mutation start coordinate in the query;
• end: mutation end coordinate in the query;
• type: mutation type identifier
  • FS_INS for frame-shifting insertions;
  • FS_DEL for frame-shifting deletions;
  • SSMD for donor splice site mutations;
  • SSMA for acceptor splice site mutations;
  • <...>
• description: a brief description of the mutation event;
• is_masked: set to MASKED for masked mutations, NOT_MASKED for inactivating ones;
• masking_reason: if mutation is masked, contains a brief explanation for masking, otherwise a single dash (“-“);
• mut_id: an internal mutation identifier.

## entries for projection XM_011511769.3#RGPD6#4039,772:
projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_011524058.3#USP6#955358	7	193	181	chr7	141187827	141187828	FS_INS	1	MASKED	Exon is missing	?FS_1
XM_011524058.3#USP6#955358	9	266	242	chr7	141188343	141188344	FS_INS	1	MASKED	Exon is deleted	FS_2
XM_011524058.3#USP6#955358	11	339	302	chr7	141189338	141189339	FS_DEL	-1	NOT_MASKED	-	?FS_3
XM_011524058.3#USP6#864	12	346	309	chr11	141189358	141189359	FS_DEL	-2	MASKED	Exon is missing FS_4
XM_011524058.3#USP6#864	12	382	344	chr11	85032116	85032117	FS_DEL	-1	MASKED	Exon is missing	FS_5
XM_011524058.3#USP6#864	13	508	469	chr11	85031669	85031670	FS_DEL	-2	MASKED	Exon is missing	FS_6

The TOGA2-generated BigBed file for the query genome browser annotation track contains information for an SVG plot visualising the reference transcript’s exon-intron structure together with the position of mutations in the query. This plot is shown when clicking on an annotated transcript in the query browser.

Mutation plots for fragmented projections are split into chain-wise subplots, each depicting the exons aligned via the respective chain:

Inactivating mutations are shown at the reference codon position. Masked mutations are plotted in pale grey. Inactivating (non-masked) mutations are plotted in red (for frame-shifting indels and splice site mutations) or black (for premature stop codons). Compensating frame-shifts are connected with green arcs.

TOGA2 also produces a separate file with decorators that can be added to the main TOGA2 BigBed browser track and visualises point mutations. The colour notation corresponds to that in the mutation files:

• inactivating mutations are coloured deep red (RGB 198,0,9);
• masked mutations are coloured pale grey (RGB 211,211,211)

The following glyph shapes are used to indicate specific mutation types (see respective mutation sections below for examples):

• triangle: frame-shifting deletions
• inverted triangle: frame-shifting insertions
• five-pointed star: premature stop codons
• square: splice site mutations

Insertions and deletions (indels) that are not divisible by three shift and thereby disrupt the reading frame in the query sequence. Frame-shifting indels are inferred from the CESAR2 alignments:

## frame-shifting insertion: alignment gap in the reference indivisible by three
ref   ATG TC-T TAG
query ATG TCTG TAG

## frame-shifting deletion: alignment gap in the query indivisible by three
ref   ATG TCT TAG
query ATG T-T TAG

In the inactivating_mutations.tsv file, frame-shifting indel entries have the following properties:

• the description column contains the size of the frame-shifting indel in nucleotides; the values are positive for insertion events (e.g., 1) or negative for deletion events (e.g., -1)
• in the mut_id column, mutation identifiers have the FS_ followed by the number of the frame-shifting mutation in the query

In the mutation plots, frame-shifting insertions are indicated by inverted triangles pointing at the insertion, and frame-shifting deletions are represented as vertical red lines. Indel size in nucleotides is shown above the symbol. Masked frameshifts are colored grey:

In the browser decorator tracks, frame-shifting indels are represented by triangle glyphs, upward for deletions and inverted for insertions:

Two or more frameshifts compensate each other if their cumulative length is zero or divisible by three and if the region in between them is translatable in the alternative frame (meaning it contains no stop codons). Because such compensating frameshifts can provide novel sequence in an otherwise intact gene (Biba et al., 2022), TOGA2 masks compensating frameshifts and does not treat them as inactivating mutations. However, compensating frameshifts can also be an indication of gene loss, when they occur together with other inactivating mutations, or nearly the entire gene is translated in a different reading frame.

Compensation events inferred by TOGA2 can include any number of insertions and deletions, except for whole exon deletions. They may lie in arbitrary distance from each other and can be located in different query sequence exons. Toy examples illustrating compensating frameshifts:

## compensated frameshift example 1: 
## two deletion events with total deletion length of -3
ref    ATG TGT GCC AGT TCA TAA
query  ATG TGC -CC GAT --A TAA
query* ATG TGC CCG ATA     TAA

## compensated frameshift example 2: 
## two indel events, an insertion and a deletion, with a total indel length of zero
ref    ATG TGT GCC AGT TC-A TAA
query  ATG TGC -CC GAT TCGA TAA
query* ATG TGC CCG ATT CGA  TAA

Notably, if a stop codon is found in the shifted frame between two compensating frameshift candidates, the mutations are not compensating and TOGA2 treats them as inactivating, as exemplified here:

## a false compensation candidate example:
## an alternative frame fragment between the two frame-shifts
## contains an in-frame TAA stop codon (lowercase symbols)
ref    ATG TGT GCC AGT TCA- TAA
query  ATG TG- Gta aGT TCCA TAA
query* ATG TGG taa GTT CCA  TAA

Because compensating frameshifts can also be an indication of gene loss, TOGA2 considers them when for the gene loss classification in two scenarios:

• If a query sequence contains non-masked inactivating mutations, such as non-masked frame-shifts, in-frame stop codons, splice site mutations, or frame-shifting exon deletions, any compensating frame-shifts are unmasked and treated as regular inactivating mutations. The rationale for this is that the more frameshifts occurred, the higher is the chance that some of them will compensate each other. Unmasked compensating frameshifts are shown in red font; however, the compensating events are still reported in the mutation files and mutation plots.
• Otherwise, the deleterious effect of alternative frame fragments is assessed at the coding sequence integrity level, where alternative frame codons are counted as Deleted towards certain integrity features’ computation.

In the inactivating_mutations.tsv file, compensating frameshifts have the MASKED label in the 10th column, with the masking_reason field set to Compensated. The compensation event entries have the following properties:

• the type field contains the COMPENSATION label
• the description field contains the mutation identifiers of the first and the last frame-shift event participating in the alternative frame, e.g., FS_1-3
• the is_masked field is set to NOT_MASKED by default to comply with the TOGA1 notation
• ff frame-shifts participating in the alternative frame formation were unmasked (see above), the masking_reason field contains the Treated as inactivating description; otherwise, the field contains a single dash
• the mut_id value contains a C_ prefix followed by the compensation number in the projection

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_047442600.1#CCNB3#26	4	112	112	chrX	7020022	7020023	COMPENSATION	FS_1-2	NOT_MASKED	-	C_1
XM_047418797.1#AFDN#3362	31	1881	1717	chr17	13975473	13975474	COMPENSATION	FS_18-19	NOT_MASKED	Treated as inactivating	C_1

In the mutation plots, compensation events are represented in grey font connected by green arcs:

Compensating frameshifts are not represented in the genome browser decoration tracks.

Mutations that create premature stop codons (TAA, TAG, and TGA) lead to a truncated protein.

## in-frame stop codon mutation: a single-base substitution in codon leads
## to a premature TGA stop codon emergence in the query reading frame
## asterisk ('*') indicates translation stop

ref   ATG CGA GGT AAC TGA
       M   R   G   N   *
query ATG TGA GGT AAC TGA
       M   *   G   N   *

In the inactivating_mutations.tsv file, premature stop codons have the following properties:

• label STOP in the type column
• the reference codon to query stop codon transition in the description column
• a STOP_ prefix in the mutation identifier in the mut_id column

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
NM_001256580.2#RPL10#286934	2	8	8	chr11	53366336	53366337	STOP	TGT->TAG	NOT_MASKED	-	STOP_1

In the mutation plots, premature stop codons are represented by a single vertical line, with the stop codon sequence shown above:

In the browser decorator tracks, premature stop codons are represented by a star glyph:

In addition to the general masking scenarios listed above, stop codons in the query are also masked if they correspond to in-frame stop codons already present in the reference sequence, where such codons can encode the 21st amino acid selenocysteine or translational stop codon readthrough events. TOGA2 masks all in-frame stop codons in the reference sequence beforehand with the NNN symbols, allowing CESAR2 to align them to stop codons in the query. Afterwards, any stop codons in the query aligned to NNN reference codons are not considered to be inactivating and are not listed in the inactivating mutation file. Instead, their sequences and coordinates are reported in the selenocysteine_codons.tsv file in the meta folder. We note that TOGA2 cannot distinguish between selenocysteine codons and stop codon readthrough.

U2-type introns typically have donor (5’) splice sites with a consensus GT or GC dinucleotide and acceptor (3’) splice sites with a consensus AG dinucleotide. TOGA2 generally considers non-consensus dinucleotide variants as splice site mutations, as such mutations typically disrupt splicing and may lead to non-functional proteins.

In the inactivating_mutations.tsv file, splice site mutations have the following features:

• the type column is set to either SSMA for acceptor site mutations or SSMD for donor site mutations
• likewise, the mut_id column identifiers start with SSMA_ for acceptor sites or SSMD_ for donor sites
• the description column contains reference and query splice site dinucleotides, separated by an arrow symbol: ag->gg

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
ENST00000222286.9#GAPDHS#343523	6	186	181	chr12	71846964	71846966	SSMA	ag->gg	NOT_MASKED	-	SSMA_3
NM_001037738.3#NPM1#172142	5	166	153	chr17	68699825	68699827	SSMD	gt->aa	NOT_MASKED	-	SSMD_4

In the mutation plots, splice site mutations are represented by splice site dinucleotide labels, written in lowercase, next to the respective intron-exon boundaries:

In the UCSC decorator track, splice site mutations are represented with squares at the respective intron-exon boundaries:

Splice site mutations are masked in the following cases:

1. Reference introns classified as U12 or non-canonical U2 can have non-GT/GC donors and non-AG acceptors with a large variety of dinucleotides. Since the intron type is often conserved, TOGA2 allows any dinucleotide to occur at such splice sites without considering them as inactivating. Such splice sites are reported in the mutation file as MASKED with the U12 intron masking reason for the reference U12 introns and Non-canonical U2 intron for non-canonical U2 introns, respectively.
2. Precise intron deletions in the query result in deleted splice site dinucleotides, but because the entire intron is deleted, such events are not considered as inactivating mutations. The events are reported in the mutation file, with Intron deletion as masking reason.
3. The splice site nucleotide contains non-standard nucleotide symbol(s). In the most common case, such exons lie directly adjacent to an assembly gap consisting of “N” symbols. The respective splice site mutations have Ambiguous symbol in the masking reason column.
4. Splice site dinucleotide lies outside of the alignment search space. Whenever possible, search spaces provided to CESAR2 contain enough context for splice site shift search. It is still possible, however, that CESAR2 aligns the exon directly adjacent to the query search space boundary, such that the full splice site dinucleotide cannot be inferred from the CESAR2 input sequence. Such mutations have the Insufficient data masking reason and question marks (“?”) for missing dinucleotide symbols in the description column:

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_047439131.1#ZNF492#522783	2	12	12	chr13	67825716	67825718	SSMA	ag->??	MASKED	Insufficient data	SSMA_1

5. In addition, all general masking scenarios (N- and C-terminal 10% of the sequence, missing and deleted exons, etc.) also apply to splice site mutations.

Large framepreserving indels can have a significant effect on the protein function. TOGA2 considers frame-preserving indels of 50 nucleotides (~17 codons) or larger as “big” indels.

In the mutation file, big indels have:

• BIG_INS (for insertions) or BIG_DEL (for deletions) in the type column
• BI_ prefix to their identifiers in the mut_id column
• indel size (+{number} for insertions and -{number} for deletions) in the description column

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_047418795.1#AFDN#3362	17_17	828_863	728_728	chr17_chr17	14054109	14054214	BIG_INS	+105	MASKED	-	BI_1
XM_047418795.1#AFDN#3362	29_29	1615_1635	1440_1460	chr17_chr17	13999007	13999009	BIG_DEL	-60	MASKED	-	BI_2

Big indels are not reflected in the mutation plots or decorator tracks, but entire exon deletions are shown in the mutation plots.

Because large length variations have been observed in conserved genes, all big indel mutations are masked by default and are not considered when classifying projections as FI. However, they might have a significant impact on the coding sequence integrity properties.Because large length variations have been observed in conserved genes, all big indel mutations are masked by default and are not considered when classifying projections as FI. However, they might have a significant impact on the coding sequence integrity properties.

By default, coding sequences are expected to start with a methionine start codon (ATG). If a reference transcript starts with ATG but the query projection contains a different codon or has its start codon deleted, TOGA2 reports a missing start codon mutation.

In the mutation file, missing start codon mutations have:

• START_MISSING in the type column
• ATG->{query_codon} in the description column, where {query_codon} is the codon found in the query; deleted start codons are reported as triple dash ("---")
• START_1 identifier in the mut_id column

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
NM_001165881.3#ZNF268#36274	1	1	1	chr2	176073672	176073673	START_MISSING	ATG->NNN	MASKED	Missing start masked	START_1

Because missing start codons may be caused by imprecise first exon annotation, especially in the case of short terminal coding exons, and rare cases of alternative start codons are known, TOGA2 does not consider missing start codon mutations as inactivating and masks them by default. Notably, projections with a missing start codon cannot be classified as FI in terms of loss status.

To account for conserved non-ATG start codons, if the reference transcript starts with a codon other than ATG, TOGA2 does not report missing start codon mutation in the query.

TOGA2 recognises the three stop codons in the eukaryotic nuclear genetic code (TAA, TAG, TGA), and reports a mutation event if an annotated query sequence does not end with any of those. In the mutation file, missing stop codon entries have the following properties:

• STOP_MISSING label in the type column
• {ref_codon}->{query_codon} label in the description column; deleted stop codons are reported as triple dash ("---")
• STOP_LOSS_1 tag in the mut_id column

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
ENST00000398612.3#ZNF596#347684	5	514	505	chr9	20347539	20347540	STOP_MISSING	TAA->---	MASKED	Missing stop masked	STOP_LOSS_1

All missing stop codon mutations are masked by default and not considered to be inactivating, because it is possible that the missing stop codon occurs downstream of the provided sequence search space. Notably, projections with a missing start codon cannot be classified as FI in terms of loss status.

To account for incompletely annotated reference transcripts, if a reference transcript does not end with a known stop codon, the stop loss mutation is still reported with the following description: Non-canonical stop in reference. These mutations are ignored for the purpose of FI projection classification.

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
ENST00000377032.5#IGKV1D-12#17234	2	124	117	chr6	69927614	69927615	STOP_MISSING	CCT->TCT	MASKED	Non-canonical stop in reference	STOP_LOSS_1

TOGA2 classifies exon candidates into three categories:

• Present: exon annotations considered trustworthy and satisfying at least one of the following criteria:
  • Overlaps its alignment chain-defined locus (if it exists) by at least one base;
  • Has nucleotide identity to the reference exon of ≥45% and a Blosum score of the translated sequence of ≥20% (these parameters separate real from random alignments very well)
  • Has both of its splice sites supported by a SpliceAI probability at or above the threshold value (0.02 by default); for terminal (first or last) exons, only their single splice site is evaluated.
• Missing: Exons not satisfying any of the criteria above and either falling outside of the alignment chain’s span or having an assembly gap in their search space. Such cases often arise from genome assembly incompleteness.
• Deleted: The remaining exons are classified as deleted and not satisfying any of the criteria for the two categories above. These exons are considered to be truly absent in the query.

Missing and Deleted exon categories are treated as mutation events, affecting the loss status of the projections. Masked exon deletions (see below) still affect the classification, preventing projections from being assigned the Fully Intact status.

Missing exons affect the loss status classification by introducing the Missing (M) codons to the codon profile during the coding sequence integrity estimation. Missing exon events are never masked, and they do not trigger compensated frame-shift unmasking.

Missing exon entries have the following properties:

• Missing exon label in the type column
• Empty description (“-") in the description column
• MIS_ identifier prefix in the mut_id column

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_047418795.1#AFDN#3362	1	1_41	1_36	chr17	14142367	14142430	Missing exon	-	NOT_MASKED	-	MIS_1

In the mutation plots, missing exon blocks are coloured grey:

Because missing exons are not annotated, they do not appear in the BED files, the BigBed browser track, and the browser decorator files.

Unlike Missing exons, exons classified as Deleted are assumed to be absent in the query genome. The impact of exon deletion on the projection loss status varies depending on its effect on the query reading frame:

• The deletion of one or more consecutive exons shifts the reading frame if their cumulative length is not divisible by three or if their deletion leads to a new in-frame stop codon between the neighbouring Present exons; such exon deletions are treated as inactivating mutations:

## case 1: exon deletion introduces a frame shift
sequence   ATG G | TT ACT AGC | TCA TGA
exon num     1   |      2     |    3
presence     I   |      D     |    I
resulting sequence: ATG G-- TCA TGA

## case 2: exon deletion introduces a premature stop codon (TGA)
sequence   ATG T | TT ACT AGC  G| AG TCA TGA
exon num     1   |      2       |     3
presence     I   |      D       |     I
resulting sequence: ATG TAG TCA TGA

• The deletion of one or more consecutive exons is frame-preserving if the cumulative exon length is divisible by three and the deletion does not lead to a split stop codon emergence. Frame-preserving exon deletions are masked.

## case 1: a single exon deletion that does not affect the frame
sequence   ATG G | TT ACT A | GC TCA TGA
exon num     1   |     2    |     3
presence     I   |     D    |     I
resulting sequence: ATG GGC TCA TGA

## case 2: two neighbouring exons
## while none of them meets the requirements above, 
## their simultaneous deletion does not shift the reading frame
sequence   ATG G | TT ACT AGC | CCA G | CT TCA TGA
exon num     1   |      2     |   3   |     4
presence     I   |      D     |   D   |     I
resulting sequence: ATG GCT TCA TGA

Deleted exon entries have the following properties:

• Deleted exon label in the type column
• Empty description - in the description column
• DEL_ identifier prefix in the DEL_ column
• MASKED label in the is_masked column if the deletion is frame-preserving, NOT_MASKED otherwise

projection	exon	triplet	ref_codon	chrom	start	end	type	description	is_masked	masking_reason	mut_id
XM_047418795.1#AFDN#3362	32	2008_2043	1821_1849	chr17	13973405	13973490	Deleted exon	-	NOT_MASKED	-	DEL_1
ENST00000421025.3#ZNF679#422068	1	1_18	1_14	chr7	42825399	42825450	Deleted exon	-	MASKED	Frame-preserving deletion	DEL_1

Frame-preserving deleted exons are colored blue in the mutation plots, and frame-shifting deleted exons are coloured red:

Because deleted exons are not annotated, they do not appear in the BED files, the BigBed browser tracks, and the decorator tracks.

Introns that are absent in the reference and annotated in the query with the help of SpliceAI do not count as inactivating mutations and do not count towards projection loss classification. Because such events indicate evolutionary exon-intron structure changes, TOGA2 reports them in the inactivating_mutations.tsv file with the following distinct features:

• column 2 (exon) contains the number of the reference exon in which the reference-absent intron was annotated;
• INTRON_GAIN label in the type column;
• single dash (“-") in the description field
• reference-absent intron entries are always MASKED, with the Intron gain masked reason in the masking_reason field

Reference-absent introns do not appear in the mutation plots. In the BED files and genome browser tracks, they appear as regular introns.