package bio_io

  1. Overview
  2. Docs

Parsing CIGAR (Concise Idiosyncratic Gapped Alignment Report) strings

Overview

A CIGAR string is a compressed representation of an aligment.

Here is an example alignment of two sequences along wth the operations, Match (M), Insertion (I), and Deletion (D).

target:     AGTG-TCTTTG
query:      ACTGAACT--G
operation:  MMMMIMMMDDM

Alignment columns with bases/residues in both the query and the target are considered a Match (even if there are two different bases/residues in the column). Columns with a gap in the target/reference sequence are labeled Insertions, and columns with a gap in the query/read sequence are labeled Deletions.

The CIGAR string is formed by counting up the runs of operations of the same type. For the above alignment, the operations are MMMMIMMMDDM, so the CIGAR string will be 4M1I3M2D1M.

Supported CIGAR Operations

  • Match: A non-gap alignment position. May be a mismatch.
  • Insertion: Gap in the target/reference sequence.
  • Deletion: Gap in the query/read sequence.

Many programs use CIGAR strings including samtools, usearch, and MMseqs2. CIGAR strings can vary depending on the program that produces or consumes them, and not all versions are compatible.

This implementation uses a reduced set of operations used by MMseqs2: Match (M), Insertion (I), and Deletion (D). Eventually, I will implement the rest of the operations, but in the meantime, if your use case requires the full CIGAR spec, please open an issue on GitHub and let me know!

Notes

  • I refer to each number-letter pair as a chunk. Each chunk is a length and an operation. The length specifies the number of consecutive operations.
  • Some programs generate chunks without a length--just the operation. In these cases, the length is taken to be 1. So the following CIGAR strings will parse into the same data structure: MIDM and 1M1I1D1M.
  • Empty CIGAR strings are allowed and do not raise errors.

API

exception Exn of Base.string
exception Int_overflow of Base.int * Base.int
type t
include Ppx_compare_lib.Equal.S with type t := t
val equal : t -> t -> bool
include Sexplib0.Sexpable.S with type t := t
val t_of_sexp : Sexplib0.Sexp.t -> t
val sexp_of_t : t -> Sexplib0.Sexp.t
val of_string_exn : Base.string -> t

of_string_exn string parses string into a cigar t. Raises exceptions on parse error. Valid cigar strings look something like this: 1M2I3D5M.

val of_string : Base.string -> t Base.Or_error.t

of_string string parses string into a cigar t Or_error.t. Should not raise exceptions. If of_string raises an exception, then you've found a bug. Valid cigar strings look something like this: 1M2I3D5M.

val to_string : t -> Base.string

to_string t converts a cigar from the internal representation to a printable string.

let s = "1M2I3D5M" in
assert (s = Cigar.to_string @@ Cigar.of_string_exn "1M2I3D5M")

Note that not all cigar strings will "round-trip" like the above example. If the input cigar string has no number specifying the length of a chunk, then that chunk is inferred to be of length 1. E.g.,

let s = "MDIM" in
assert ("1M1D1I1M" = Cigar.to_string @@ Cigar.of_string_exn "1M2I3D5M")

Length & count functions

The following functions return information about the aligment as inferred from the CIGAR string. Each of the examples uses this CIGAR string, 1M2I3D5M, which represents the following alignment.

target: X--XXXXXXXX
query:  XXX---XXXXX
op:     MIIDDDMMMMM

Then "length" calculating functions can overflow on CIGAR strings describing genome alignments. In practice, this won't happen often as generally you will be parsing the results of local alignments, e.g., from BLAST, MMseqs2, or reading SAM files.

Since it is possible to overflow integer arithmetic on CIGAR strings you may parse, the length functions use an addition function that raises Int_overflow exceptions rather than use two's complement wrapping.

Thus, there are two versions, those that end in _exn that may raise Int_overflow and those that return t Or_error.t, which catch Int_overflow and other exceptions.

val alignment_length_exn : t -> Base.int
val alignment_length : t -> Base.int Base.Or_error.t

alignment_length t returns the total length of the aligment as inferred by the Cigar string.

Here is the alignment for this CIGAR string:

assert (11 = Cigar.alignment_length @@ Cigar.of_string_exn "1M2I3D5M")
val num_gaps_exn : t -> Base.int
val num_gaps : t -> Base.int Base.Or_error.t

num_gaps t returns the total number of gap columns in the alignment.

assert (5 = Cigar.num_gaps @@ Cigar.of_string_exn "1M2I3D5M")
val num_matches_exn : t -> Base.int
val num_matches : t -> Base.int Base.Or_error.t

num_matches t returns the number of matches in the alignment.

assert (6 = Cigar.num_matches @@ Cigar.of_string_exn "1M2I3D5M")

This is "ungapped alignment length" as in https://doi.org/10.1093/bioinformatics/bty262. Where ungapped is the alignment length minus number of gaps. I'm taking that to mean any position with a gap in either sequence is not counted. So the ungapped_length is the number of matches.

val query_length_exn : t -> Base.int
val query_length : t -> Base.int Base.Or_error.t

query_length t returns the length of the query/read within the boundaries of the alignment. This is the length of the query as inferred by the Cigar string. If it is a local aligment, it may not be equal to the length of the entire query sequence.

assert (8 = Cigar.query_length @@ Cigar.of_string_exn "1M2I3D5M")
val target_length_exn : t -> Base.int
val target_length : t -> Base.int Base.Or_error.t

target_length t returns the length of the target/reference within the boundaries of the alignment. This is the length of the target as inferred by the Cigar string. If it is a local aligment, it may not be equal to the length of the entire target sequence.

assert (9 = Cigar.target_length @@ Cigar.of_string_exn "1M2I3D5M")

Drawing Functions

Drawing functions can also raise Int_overflow because they rely on the length functions. Use draw rather than draw_exn to catch errors.

val draw_exn : ?max_aln_len:Base.int -> ?gap:Base.char -> ?non_gap:Base.char -> ?wrap:Base.int -> t -> Base.string
val draw : ?max_aln_len:Base.int -> ?gap:Base.char -> ?non_gap:Base.char -> ?wrap:Base.int -> t -> Base.string Base.Or_error.t

draw t ~max_aln_len ~gap ~non_gap ~wrap outputs a string representation of the aligmnent as inferred from the CIGAR string. This function is mainly just for fun or if you want to learn how CIGAR strings work :)

  • Use max_aln_len to prevent "drawing" alignments that are longer than this value. Currently, this function allocates a string approx. three times the entire size of the alignment. So keep this low... :) Default is 1000.
  • gap is the character to use for gaps. Default: '-'.
  • non_gap is the character to use for non-gaps. Default 'X'.
  • wrap is the max length of the sequence to show in a line before wrapping. Default is 60.

E.g., this print_endline @@ draw "1M2I3D5M"

would print this:

target: X--XXXXXXXX
query:  XXX---XXXXX
op:     MIIDDDMMMMM

Wrapping long sequences

You can wrap long sequences so they don't take up so much horizontal space. This code

print_endline @@ Cigar.draw ~wrap:10 @@ Cigar.of_string_exn "25M"

will print out

t: XXXXXXXXXX
q: XXXXXXXXXX
o: MMMMMMMMMM

t: XXXXXXXXXX
q: XXXXXXXXXX
o: MMMMMMMMMM

t: XXXXX
q: XXXXX
o: MMMMM
OCaml

Innovation. Community. Security.