Skip to content

DnaTokenizer

DnaTokenizer is smart, it tokenizes raw DNA nucleotides into tokens, no matter if the input is in uppercase or lowercase, uses T (Thymine) or U (Uracil), and with or without special tokens. It also supports tokenization into nmers and codons, so you don’t have to write complex code to preprocess your data.

By default, DnaTokenizer uses the standard alphabet. If nmers is greater than 1, or codon is set to True, it will instead use the streamline alphabet.

multimolecule.tokenisers.DnaTokenizer

Bases: Tokenizer

Tokenizer for DNA sequences.

Parameters:

Name Type Description Default
alphabet Alphabet | str | List[str] | None

alphabet to use for tokenization.

  • If is None, the standard RNA alphabet will be used.
  • If is a string, it should correspond to the name of a predefined alphabet. The options include
    • standard
    • iupac
    • streamline
    • nucleobase
  • If is an alphabet or a list of characters, that specific alphabet will be used.
 None
nmers int

Size of kmer to tokenize.

 1
codon bool

Whether to tokenize into codons.

 False
replace_U_with_T bool

Whether to replace U with T.

 True
do_upper_case bool

Whether to convert input to uppercase.

 True

Examples:

Python Console Session
>>> from multimolecule import DnaTokenizer
>>> tokenizer = DnaTokenizer()
>>> tokenizer('<pad><cls><eos><unk><mask><null>ACGTNRYSWKMBDHV.X*-')["input_ids"]
[1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 2]
>>> tokenizer('acgt')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer('acgu')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer = DnaTokenizer(replace_U_with_T=False)
>>> tokenizer('acgu')["input_ids"]
[1, 6, 7, 8, 3, 2]
>>> tokenizer = DnaTokenizer(nmers=3)
>>> tokenizer('tataaagta')["input_ids"]
[1, 84, 21, 81, 6, 8, 19, 71, 2]
>>> tokenizer = DnaTokenizer(codon=True)
>>> tokenizer('tataaagta')["input_ids"]
[1, 84, 6, 71, 2]
>>> tokenizer('tataaagtaa')["input_ids"]
Traceback (most recent call last):
ValueError: length of input sequence must be a multiple of 3 for codon tokenization, but got 10
Source code in multimolecule/tokenisers/dna/tokenization_dna.py
Python
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
class DnaTokenizer(Tokenizer):
    """
    Tokenizer for DNA sequences.

    Args:
        alphabet: alphabet to use for tokenization.

            - If is `None`, the standard RNA alphabet will be used.
            - If is a `string`, it should correspond to the name of a predefined alphabet. The options include
                + `standard`
                + `iupac`
                + `streamline`
                + `nucleobase`
            - If is an alphabet or a list of characters, that specific alphabet will be used.
        nmers: Size of kmer to tokenize.
        codon: Whether to tokenize into codons.
        replace_U_with_T: Whether to replace U with T.
        do_upper_case: Whether to convert input to uppercase.

    Examples:
        >>> from multimolecule import DnaTokenizer
        >>> tokenizer = DnaTokenizer()
        >>> tokenizer('<pad><cls><eos><unk><mask><null>ACGTNRYSWKMBDHV.X*-')["input_ids"]
        [1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 2]
        >>> tokenizer('acgt')["input_ids"]
        [1, 6, 7, 8, 9, 2]
        >>> tokenizer('acgu')["input_ids"]
        [1, 6, 7, 8, 9, 2]
        >>> tokenizer = DnaTokenizer(replace_U_with_T=False)
        >>> tokenizer('acgu')["input_ids"]
        [1, 6, 7, 8, 3, 2]
        >>> tokenizer = DnaTokenizer(nmers=3)
        >>> tokenizer('tataaagta')["input_ids"]
        [1, 84, 21, 81, 6, 8, 19, 71, 2]
        >>> tokenizer = DnaTokenizer(codon=True)
        >>> tokenizer('tataaagta')["input_ids"]
        [1, 84, 6, 71, 2]
        >>> tokenizer('tataaagtaa')["input_ids"]
        Traceback (most recent call last):
        ValueError: length of input sequence must be a multiple of 3 for codon tokenization, but got 10
    """

    model_input_names = ["input_ids", "attention_mask"]

    def __init__(
        self,
        alphabet: Alphabet | str | List[str] | None = None,
        nmers: int = 1,
        codon: bool = False,
        replace_U_with_T: bool = True,
        do_upper_case: bool = True,
        additional_special_tokens: List | Tuple | None = None,
        **kwargs,
    ):
        if codon and (nmers > 1 and nmers != 3):
            raise ValueError("Codon and nmers cannot be used together.")
        if codon:
            nmers = 3  # set to 3 to get correct vocab
        if not isinstance(alphabet, Alphabet):
            alphabet = get_alphabet(alphabet, nmers=nmers)
        super().__init__(
            alphabet=alphabet,
            nmers=nmers,
            codon=codon,
            replace_U_with_T=replace_U_with_T,
            do_upper_case=do_upper_case,
            additional_special_tokens=additional_special_tokens,
            **kwargs,
        )
        self.replace_U_with_T = replace_U_with_T
        self.nmers = nmers
        self.condon = codon

    def _tokenize(self, text: str, **kwargs):
        if self.do_upper_case:
            text = text.upper()
        if self.replace_U_with_T:
            text = text.replace("U", "T")
        if self.condon:
            if len(text) % 3 != 0:
                raise ValueError(
                    f"length of input sequence must be a multiple of 3 for codon tokenization, but got {len(text)}"
                )
            return [text[i : i + 3] for i in range(0, len(text), 3)]
        if self.nmers > 1:
            return [text[i : i + self.nmers] for i in range(len(text) - self.nmers + 1)]  # noqa: E203
        return list(text)

MultiMolecule provides a set of predefined alphabets for tokenization.

Standard Alphabet

The standard alphabet is an extended version of the IUPAC alphabet. This extension includes two additional symbols to the IUPAC alphabet, X and *.

  • X: Any base; is slightly different from N which represents Unknown base. In automatic word embedding conversion, the X will be initialized as the mean of A, C, G, and T, while N will not be further processed.
  • *: is not used in MultiMolecule and is reserved for future use.

gap

Note that we use . to represent a gap in the sequence.

While - exists in the standard alphabet, it is not used in MultiMolecule and is reserved for future use.

Code Represents
A Adenine
C Cytosine
G Guanine
T Thymine
N Unknown
R A or G
Y C or T
S C or G
W A or T
K G or T
M A or C
B C, G, or T
D A, G, or T
H A, C, or T
V A, C, or G
. Gap
X Any
* Not Used
- Not Used

IUPAC Alphabet

IUPAC nucleotide code is a standard nucleotide code proposed by the International Union of Pure and Applied Chemistry (IUPAC) to represent DNA sequences.

It consists of 10 symbols that represent ambiguity in the nucleotide sequence and 1 symbol that represents a gap in addition to the streamline alphabet.

Code Represents
A Adenine
C Cytosine
G Guanine
T Thymine
R A or G
Y C or T
S C or G
W A or T
K G or T
M A or C
B C, G, or T
D A, G, or T
H A, C, or T
V A, C, or G
N A, C, G, or T
. Gap

Note that we use . to represent a gap in the sequence.

Streamline Alphabet

The streamline alphabet includes one additional symbol to the nucleobase alphabet, N to represent unknown nucleobase.

Code Nucleotide
A Adenine
C Cytosine
G Guanine
T Thymine
N Unknown

Nucleobase Alphabet

The nucleobase alphabet is a minimal version of the DNA alphabet that includes only the four canonical nucleotides A, C, G, and T.

Code Nucleotide
A Adenine
C Cytosine
G Guanine
T Thymine