Skip to content

Embedding Models

taxotagger.models

ModelFactory

Factory class to get the embedding model for the given model identifier.

get_model staticmethod 

get_model(
    model_id: str, config: ProjectConfig
) -> EmbedModelBase

Get the embedding model for the given model identifier.

Parameters:

  • model_id (str) –

    The identifier of the model to load.

  • config (ProjectConfig) –

    The configurations for the project.

Returns:

  • EmbedModelBase –

    The embedding model instance for the given model identifier.

Examples:

>>> config = ProjectConfig()
>>> model = ModelFactory.get_model("MycoAI-CNN", config)
Source code in src/taxotagger/models.py 
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
@staticmethod
def get_model(model_id: str, config: ProjectConfig) -> EmbedModelBase:
    """Get the embedding model for the given model identifier.

    Args:
        model_id: The identifier of the model to load.
        config: The configurations for the project.

    Returns:
        The embedding model instance for the given model identifier.

    Examples:
        >>> config = ProjectConfig()
        >>> model = ModelFactory.get_model("MycoAI-CNN", config)
    """
    if model_id == "MycoAI-CNN":
        return MycoAICNNEmbedModel(config)
    elif model_id == "MycoAI-BERT":
        return MycoAIBERTEmbedModel(config)
    # Add more embedding models here if needed
    else:
        raise ValueError(
            f"Invalid model id {model_id}. Valid models are {PRETRAINED_MODELS.keys()}"
        )

MycoAICNNEmbedModel 

MycoAICNNEmbedModel(config: ProjectConfig)

Bases: EmbedModelBase

Embedding model for the pretrained MycoAI-CNN.

Source code in src/taxotagger/models.py 
54
55
56
def __init__(self, config: ProjectConfig) -> None:
    self._config = config
    self.model = load_model(self.name, config)

name class-attribute instance-attribute 

name = 'MycoAI-CNN'

model instance-attribute 

model = load_model(name, config)

embed 

embed(fasta_file: str) -> dict[str, list[dict[str, Any]]]

Calculate the embeddings for the given FASTA file.

Parameters:

  • fasta_file (str) –

    The path to the FASTA file to embed.

Returns:

  • dict[str, list[dict[str, Any]]] –

    A dictionary of embeddings for each taxonomy level. The dictionary keys are the taxonomy levels, and the values are lists of dictionaries containing the id, embeddings and metadata for each sequence.

    The shape of the list is (n_samples), where n_samples is the number of sequences.

    The keys of the inside dictionaries are: id, vector, and the taxonomy levels (e.g. phylum, class, order, family, genus, species) and other metadata fields present in the FASTA header.

    The shape of the vector is (n_features), where n_features is the number of features in the embedding. The number of features for each taxonomy level is:

    - phylum: 18
- class: 70
- order: 231
- family: 791
- genus: 3695
- species: 14742

    The returned data looks like: 

    {
"phylum": [{"id": "seq1", "vector": [0.1, 0.2, ...], "phylum": "Basidiomycota", ...}, ...],
"class": [{"id": "seq1", "vector": [0.5, 0.6, ...], "class": "Agaricomycetes", ...}, ...],
"order": [{"id": "seq1", "vector": [0.9, 0.8, ...], "order": "Corticiales", ...}, ...],
"family": [{"id": "seq1", "vector": [0.3, 0.4, ...], "family": "Corticiaceae", ...}, ...],
"genus": [{"id": "seq1", "vector": [0.7, 0.8, ...], "genus": "Waitea", ...}, ...],
"species": [{"id": "seq1", "vector": [0.5, 0.6, ...], "species": "Circinata", ...}, ...]
}

Examples:

>>> config = ProjectConfig()
>>> model = MycoAICNNEmbedModel(config)
>>> embeddings = model.embed("dna1.fasta")
Source code in src/taxotagger/models.py 
 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
117
118
119
120
121
122
123
124
125
126
127
128
129
def embed(self, fasta_file: str) -> dict[str, list[dict[str, Any]]]:
    """Calculate the embeddings for the given FASTA file.

    Args:
        fasta_file: The path to the FASTA file to embed.

    Returns:
        A dictionary of embeddings for each taxonomy level.
            The dictionary keys are the taxonomy levels, and the values are lists of dictionaries
            containing the id, embeddings and metadata for each sequence.

            The shape of the list is `(n_samples)`, where `n_samples` is the number of sequences.

            The keys of the inside dictionaries are: `id`, `vector`, and the taxonomy levels
            (e.g. `phylum`, `class`, `order`, `family`, `genus`, `species`) and other metadata
            fields present in the FASTA header.

            The shape of the `vector` is `(n_features)`, where `n_features` is the number of
            features in the embedding. The number of features for each taxonomy level is:

                - phylum: 18
                - class: 70
                - order: 231
                - family: 791
                - genus: 3695
                - species: 14742

            The returned data looks like:
            ```python
            {
            "phylum": [{"id": "seq1", "vector": [0.1, 0.2, ...], "phylum": "Basidiomycota", ...}, ...],
            "class": [{"id": "seq1", "vector": [0.5, 0.6, ...], "class": "Agaricomycetes", ...}, ...],
            "order": [{"id": "seq1", "vector": [0.9, 0.8, ...], "order": "Corticiales", ...}, ...],
            "family": [{"id": "seq1", "vector": [0.3, 0.4, ...], "family": "Corticiaceae", ...}, ...],
            "genus": [{"id": "seq1", "vector": [0.7, 0.8, ...], "genus": "Waitea", ...}, ...],
            "species": [{"id": "seq1", "vector": [0.5, 0.6, ...], "species": "Circinata", ...}, ...]
            }
            ```

    Examples:
        >>> config = ProjectConfig()
        >>> model = MycoAICNNEmbedModel(config)
        >>> embeddings = model.embed("dna1.fasta")
    """
    headers, encoded_data = self.parse_and_encode_fasta(fasta_file)
    # headers shape (n_samples, n_headers), e.g.
    # [['id1', 'phylum1', 'class1', 'order1', 'family1', 'genus1', 'species1', 'SH_id1'], ...]

    embeddings = []
    dataloader = DataLoader(encoded_data, shuffle=False)
    with torch.no_grad():
        for x, _ in dataloader:  # (encoded data, labels)
            y_pred = self.model(x.to(self._config.device))
            embeddings.append(y_pred)
    # embeddings shape (n_samples, n_taxonomies, (1, n_features)), where n_taxonomies is 6, e.g.
    # [[phylumTensor1, classTensor1, orderTensor1, familyTensor1, genusTensor1, speciesTensor1], ...]
    # the shape of each tensor is (1, n_features), n_features are different for each taxonomy level

    data_collections = {}
    for i, taxo_level in enumerate(TAXONOMY_LEVELS):
        data_list = [
            {
                "id": headers[j][0],
                # squeeze to remove the batch dimension: (1, n_features) -> (n_features)
                "vector": embeddings[j][i].squeeze().numpy(),
                taxo_level: headers[j][i + 2],
                "SH_id": headers[j][-1],
            }
            for j in range(len(embeddings))
        ]
        data_collections[taxo_level] = data_list
    return data_collections

parse_and_encode_fasta 

parse_and_encode_fasta(
    fasta_file: str,
) -> tuple[list[list[str]], TensorData]

Parse headers and encode the sequences in the given FASTA file.

The sequences are encoded using the encoders defined in the pretrained model.

Parameters:

  • fasta_file (str) –

    The path to the FASTA file.

Returns:

  • tuple[list[list[str]], TensorData] –

    A tuple containing the headers and the encoded data for the sequences in the FASTA file.

    The shape of the headers is (n_samples, n_headers), where n_samples is the number of sequences and n_headers is the 9 metadata fields parsed from the header.

Source code in src/taxotagger/models.py 
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
def parse_and_encode_fasta(self, fasta_file: str) -> tuple[list[list[str]], data.TensorData]:
    """Parse headers and encode the sequences in the given FASTA file.

    The sequences are encoded using the encoders defined in the pretrained model.

    Args:
        fasta_file: The path to the FASTA file.

    Returns:
        A tuple containing the headers and the encoded data for the sequences in the FASTA file.

            The shape of the headers is `(n_samples, n_headers)`, where `n_samples` is the
            number of sequences and `n_headers` is the 9 metadata fields parsed from the header.
    """
    input_data = data.Data(fasta_file, tax_parser=None, allow_duplicates=False)
    # Using custom parser to parse the FASTA headers
    headers = [parse_unite_fasta_header(header) for header in input_data.data["id"].values]
    encoded_data = input_data.encode_dataset(self.model.dna_encoder, self.model.tax_encoder)
    return headers, encoded_data

MycoAIBERTEmbedModel 

MycoAIBERTEmbedModel(config: ProjectConfig)

Bases: EmbedModelBase

Embedding model for the pretrained MycoAI-BERT.

Source code in src/taxotagger/models.py 
157
158
159
def __init__(self, config: ProjectConfig) -> None:
    self._config = config
    self.model = load_model(self.name, config)

name class-attribute instance-attribute 

name = 'MycoAI-BERT'

model instance-attribute 

model = load_model(name, config)

embed 

embed(fasta_file: str) -> dict[str, list[dict[str, Any]]]

Calculate the embeddings for the given FASTA file.

Parameters:

  • fasta_file (str) –

    The path to the FASTA file to embed.

Returns:

  • dict[str, list[dict[str, Any]]] –

    A dictionary of embeddings for each taxonomy level. The dictionary keys are the taxonomy levels, and the values are lists of dictionaries containing the id, embeddings and metadata for each sequence.

    The shape of the list is (n_samples), where n_samples is the number of sequences.

    The keys of the inside dictionaries are: id, vector, and the taxonomy levels (e.g. phylum, class, order, family, genus, species) and other metadata fields present in the FASTA header.

    The shape of the vector is (n_features), where n_features is the number of features in the embedding. The number of features for each taxonomy level is:

    - phylum: 18
- class: 70
- order: 231
- family: 791
- genus: 3695
- species: 14742

    The returned data looks like: 

    {
"phylum": [{"id": "seq1", "vector": [0.1, 0.2, ...], "phylum": "Basidiomycota", ...}, ...],
"class": [{"id": "seq1", "vector": [0.5, 0.6, ...], "class": "Agaricomycetes", ...}, ...],
"order": [{"id": "seq1", "vector": [0.9, 0.8, ...], "order": "Corticiales", ...}, ...],
"family": [{"id": "seq1", "vector": [0.3, 0.4, ...], "family": "Corticiaceae", ...}, ...],
"genus": [{"id": "seq1", "vector": [0.7, 0.8, ...], "genus": "Waitea", ...}, ...],
"species": [{"id": "seq1", "vector": [0.5, 0.6, ...], "species": "Circinata", ...}, ...]
}

Examples:

>>> config = ProjectConfig()
>>> model = MycoAIBERTEmbedModel(config)
>>> embeddings = model.embed("dna1.fasta")
Source code in src/taxotagger/models.py 
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
def embed(self, fasta_file: str) -> dict[str, list[dict[str, Any]]]:
    """Calculate the embeddings for the given FASTA file.

    Args:
        fasta_file: The path to the FASTA file to embed.

    Returns:
        A dictionary of embeddings for each taxonomy level.
            The dictionary keys are the taxonomy levels, and the values are lists of dictionaries
            containing the id, embeddings and metadata for each sequence.

            The shape of the list is `(n_samples)`, where `n_samples` is the number of sequences.

            The keys of the inside dictionaries are: `id`, `vector`, and the taxonomy levels
            (e.g. `phylum`, `class`, `order`, `family`, `genus`, `species`) and other metadata
            fields present in the FASTA header.

            The shape of the `vector` is `(n_features)`, where `n_features` is the number of
            features in the embedding. The number of features for each taxonomy level is:

                - phylum: 18
                - class: 70
                - order: 231
                - family: 791
                - genus: 3695
                - species: 14742

            The returned data looks like:
            ```python
            {
            "phylum": [{"id": "seq1", "vector": [0.1, 0.2, ...], "phylum": "Basidiomycota", ...}, ...],
            "class": [{"id": "seq1", "vector": [0.5, 0.6, ...], "class": "Agaricomycetes", ...}, ...],
            "order": [{"id": "seq1", "vector": [0.9, 0.8, ...], "order": "Corticiales", ...}, ...],
            "family": [{"id": "seq1", "vector": [0.3, 0.4, ...], "family": "Corticiaceae", ...}, ...],
            "genus": [{"id": "seq1", "vector": [0.7, 0.8, ...], "genus": "Waitea", ...}, ...],
            "species": [{"id": "seq1", "vector": [0.5, 0.6, ...], "species": "Circinata", ...}, ...]
            }
            ```

    Examples:
        >>> config = ProjectConfig()
        >>> model = MycoAIBERTEmbedModel(config)
        >>> embeddings = model.embed("dna1.fasta")
    """
    headers, encoded_data = self.parse_and_encode_fasta(fasta_file)
    # headers shape (n_samples, n_headers), e.g.
    # [['id1', 'phylum1', 'class1', 'order1', 'family1', 'genus1', 'species1', 'SH_id1'], ...]

    embeddings = []
    dataloader = DataLoader(encoded_data, shuffle=False)
    with torch.no_grad():
        for x, _ in dataloader:  # (encoded data, labels)
            y_pred = self.model(x.to(self._config.device))
            embeddings.append(y_pred)
    # embeddings shape (n_samples, n_taxonomies, (1, n_features)), where n_taxonomies is 6, e.g.
    # [[phylumTensor1, classTensor1, orderTensor1, familyTensor1, genusTensor1, speciesTensor1], ...]
    # the shape of each tensor is (1, n_features), n_features are different for each taxonomy level

    data_collections = {}
    for i, taxo_level in enumerate(TAXONOMY_LEVELS):
        data_list = [
            {
                "id": headers[j][0],
                # squeeze to remove the batch dimension: (1, n_features) -> (n_features)
                "vector": embeddings[j][i].squeeze().numpy(),
                taxo_level: headers[j][i + 2],
                "SH_id": headers[j][-1],
            }
            for j in range(len(embeddings))
        ]
        data_collections[taxo_level] = data_list
    return data_collections

parse_and_encode_fasta 

parse_and_encode_fasta(
    fasta_file: str,
) -> tuple[list[list[str]], TensorData]

Parse headers and encode the sequences in the given FASTA file.

The sequences are encoded using the encoders defined in the pretrained model.

Parameters:

  • fasta_file (str) –

    The path to the FASTA file.

Returns:

  • tuple[list[list[str]], TensorData] –

    A tuple containing the headers and the encoded data for the sequences in the FASTA file.

    The shape of the headers is (n_samples, n_headers), where n_samples is the number of sequences and n_headers is the 9 metadata fields parsed from the header.

Source code in src/taxotagger/models.py 
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
def parse_and_encode_fasta(self, fasta_file: str) -> tuple[list[list[str]], data.TensorData]:
    """Parse headers and encode the sequences in the given FASTA file.

    The sequences are encoded using the encoders defined in the pretrained model.

    Args:
        fasta_file: The path to the FASTA file.

    Returns:
        A tuple containing the headers and the encoded data for the sequences in the FASTA file.

            The shape of the headers is `(n_samples, n_headers)`, where `n_samples` is the
            number of sequences and `n_headers` is the 9 metadata fields parsed from the header.
    """
    input_data = data.Data(fasta_file, tax_parser=None, allow_duplicates=False)
    # Using custom parser to parse the FASTA headers
    headers = [parse_unite_fasta_header(header) for header in input_data.data["id"].values]
    encoded_data = input_data.encode_dataset(self.model.dna_encoder, self.model.tax_encoder)
    return headers, encoded_data