File size: 16,742 Bytes
4fd4743
9da3822
 
 
6991a45
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
989191e
 
 
 
 
 
 
 
6991a45
 
 
 
989191e
 
 
6991a45
4fd4743
 
989191e
4fd4743
f39747b
989191e
4fd4743
 
 
 
989191e
 
 
f39747b
989191e
 
 
 
4fd4743
 
 
f39747b
4fd4743
989191e
4fd4743
989191e
 
 
 
 
4fd4743
 
989191e
 
f39747b
989191e
 
 
 
 
 
4fd4743
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
989191e
4fd4743
f39747b
4fd4743
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
989191e
 
4fd4743
989191e
 
 
 
 
 
 
 
4fd4743
 
989191e
 
f39747b
989191e
 
 
 
 
4fd4743
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
989191e
4fd4743
6991a45
 
 
989191e
4fd4743
 
989191e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fd4743
 
 
 
 
 
 
 
 
 
6991a45
 
 
 
 
 
f39747b
e610f37
 
6991a45
 
 
 
 
 
 
 
 
4fd4743
 
 
 
 
 
 
 
f39747b
4fd4743
f39747b
 
4fd4743
 
 
 
 
 
 
 
 
 
 
 
 
 
 
989191e
 
 
4fd4743
 
 
989191e
4fd4743
 
 
f39747b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
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
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
---
license: mit
language:
- en
widget:
- source_sentence: 'Ventral humeral ridge: or not'
  sentences:
  - >-
    If metasternum ossified, shape: long, narrow and tapering markedly
    anteriorly to posteriorly, length up to 3.5 times maximum width
  - >-
    Astragalus, dorsolateral margin:: overlaps the anterior and posterior
    portions of the calcaneum equally
  - 'Ulna size: does not apply'
- source_sentence: >-
    Form of distal portion of anteroventral process of ectopterygoid: varyingly
    falcate
  sentences:
  - 'Middle and distal radials in dorsal and anal fins: absent'
  - >-
    Degree of development of primitively medial portion of fourth upper
    pharyngeal tooth-plate: fourth upper pharyngeal tooth-plate covers ventral,
    posterior, dorsal and sometimes anterior surfaces of fourth
    infrapharyngobranchial
  - 'Shape of pharyngeal apophysis (basioccipital): forked anteriorly'
- source_sentence: >-
    Form of distal portion of anteroventral process of ectopterygoid: varyingly
    falcate
  sentences:
  - 'parhypural: present'
  - 'Epural: heavy'
  - 'First infraorbital: short'
- source_sentence: >-
    Form of distal portion of anteroventral process of ectopterygoid: varyingly
    falcate
  sentences:
  - 'Dentary and angular: touch'
  - 'Urohyal and first basibranchial: firmly attached'
  - 'Supraneural 3-4 (nonadditive): absent'
- source_sentence: >-
    Form of distal portion of anteroventral process of ectopterygoid: varyingly
    falcate
  sentences:
  - 'Ventral diverging lamellae of mesethmoid: lamellae reduced or absent'
  - 'Ventral ridge of the coracoid with a posterior process: absent'
  - 'carpals: fully or partially ossified'
pipeline_tag: sentence-similarity
library_name: sentence-transformers
base_model: sentence-transformers/all-mpnet-base-v2
metrics:
- pearson_cosine
- spearman_cosine
model-index:
- name: SentenceTransformer based on sentence-transformers/all-mpnet-base-v2
  results:
  - task:
      type: semantic-similarity
      name: Semantic Similarity
    dataset:
      name: pheno dev
      type: pheno-dev
    metrics:
    - type: pearson_cosine
      value: 0.6082332469417436
      name: Pearson Cosine
    - type: spearman_cosine
      value: 0.6250387873495056
      name: Spearman Cosine
  - task:
      type: semantic-similarity
      name: Semantic Similarity
    dataset:
      name: pheno test
      type: pheno-test
    metrics:
    - type: pearson_cosine
      value: 0.6822053314599665
      name: Pearson Cosine
    - type: spearman_cosine
      value: 0.705688010939619
      name: Spearman Cosine
tags:
- ontology
- nlp
- biology
- animals
- fish
- embedding
- trait
- sentence-transformers
- sentence-similarity
- feature-extraction
- loss:CoSENTLoss
datasets: 
- imageomics/char-sim-data
model_name: Trait2Vec
model_description: "Language model for embedding organismal trait descriptions. Built using Sentence-Transformer architecture and trained with trait descriptions from Imageomics/char-sim-data."
---

# Model Card for Trait2Vec

Trait2Vec is a language model to embed organismal trait descriptions in a way that preserves the structure induced by a semantic similarity metric (e.g. SimGIC). The model was trained on the [Character Similarity Dataset](https://huggingface.co/datasets/imageomics/char-sim-data). It is fine-tuned from [sentence-transformers/all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2). 
Through qualitative data exploration we observe the cosine similarity between embeddings of raw trait description is proportional to the semantic similarity of their corresponding ontological representations. 

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->

- **Developed by:** Juan Garcia, Soumyashree Kar, Jim Balhoff, Hilmar Lapp 
- **Model type:** Sentence Transformer
- **Language(s) (NLP):** English
- **License:** MIT
- **Fine-tuned from model:** [sentence-transformers/all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2)

### Model Sources

- **Repository:** [Imageomics/char-sim](https://github.com/Imageomics/char-sim/)

## Uses

Trait2Vec has been qualitatively evaluated in the ability to embed raw trait descriptions in a way that preserves the structure of an ontology. Accordingly, we expect it to produce an alternative computational representation of the traits of an organism. 

### Direct Use

It can be used to embed the textual trait descriptions associated with an organism.


## Bias, Risks, and Limitations

This model is finetuned from  [sentence-transformers/all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2), therefore it inherits its corresponding biases and risks. The training dataset([Character Similarity Dataset](https://huggingface.co/datasets/imageomics/char-sim-data)) introduces the biases of the single similarity metric and ontology. This means the embedding inherits that metric’s inductive biases, coverage gaps, and evolving definitions. Biological conclusions may differ under alternative metrics (e.g., Resnik, Jaccard) or other phenotype ontologies. 

### Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

## How to Get Started with the Model

First install the Sentence Transformers library:

```bash
pip install -U sentence-transformers
```

Then you can load this model and run inference.
```python
from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("imageomics/trait2vec")
# Run inference
sentences = [
    'Form of distal portion of anteroventral process of ectopterygoid: varyingly falcate',
    'Ventral ridge of the coracoid with a posterior process: absent',
    'carpals: fully or partially ossified',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 256]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]
```

## Training Details

### Training Data

This model was trained on the [Character Similarity Dataset](https://huggingface.co/datasets/imageomics/char-sim-data).

* Size: 438,516 training samples
* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>score</code>
* Approximate statistics based on the first 1000 samples:
  |         | sentence1                                                                          | sentence2                                                                         | score                                                          |
  |:--------|:-----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:---------------------------------------------------------------|
  | type    | string                                                                             | string                                                                            | float                                                          |
  | details | <ul><li>min: 9 tokens</li><li>mean: 42.84 tokens</li><li>max: 164 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 22.8 tokens</li><li>max: 164 tokens</li></ul> | <ul><li>min: 0.0</li><li>mean: 0.1</li><li>max: 0.61</li></ul> |
* Samples:
  | sentence1                                                                                      | sentence2                                                                                                                                                                                                    | score                             |
  |:-----------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:----------------------------------|
  | <code>Gill raker shape between ceratobranchial 1 and ceratobranchials 2--4: Homomorphic</code> | <code>Extent of development of inferior lamella of lateral ethmoid: inferior lamella absent</code>                                                                                                           | <code>0.014706667500582846</code> |
  | <code>Gill raker shape between ceratobranchial 1 and ceratobranchials 2--4: Homomorphic</code> | <code>Shape of anal-fin pterygiophore tips: tips of pterygiophores shaped like an arrow-head; axial series of pterygiophores providing the ventral margin of the anal-fin base a scalloped appearance</code> | <code>0.030538703023734296</code> |
  | <code>Gill raker shape between ceratobranchial 1 and ceratobranchials 2--4: Homomorphic</code> | <code>Suprapreopercle: present</code>                                                                                                                                                                        | <code>0.3385057414877959</code>   |
* Loss: [<code>CoSENTLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosentloss) with these parameters:
  ```json
  {
      "scale": 20.0,
      "similarity_fct": "pairwise_cos_sim"
  }
  ```

#### Training Hyperparameters
#### Non-Default Hyperparameters

- `eval_strategy`: steps
- `per_device_train_batch_size`: 64
- `per_device_eval_batch_size`: 64
- `learning_rate`: 2e-05
- `num_train_epochs`: 10
- `warmup_ratio`: 1e-06

- **Training regime:** fp32


## Evaluation

We tested Trait2Vec on a hold-out split of 20\% of the [Character Similarity Dataset](https://huggingface.co/datasets/imageomics/char-sim-data/). No descriptor overlap was ensured.

### Testing Data, Factors & Metrics

#### Testing Data

* Size: 111,628 evaluation samples
* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>score</code>
* Approximate statistics based on the first 1000 samples:
  |         | sentence1                                                                         | sentence2                                                                          | score                                                          |
  |:--------|:----------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------|:---------------------------------------------------------------|
  | type    | string                                                                            | string                                                                             | float                                                          |
  | details | <ul><li>min: 9 tokens</li><li>mean: 17.19 tokens</li><li>max: 24 tokens</li></ul> | <ul><li>min: 3 tokens</li><li>mean: 21.97 tokens</li><li>max: 143 tokens</li></ul> | <ul><li>min: 0.0</li><li>mean: 0.1</li><li>max: 0.86</li></ul> |
* Samples:
  | sentence1                                  | sentence2                                                                                                         | score                             |
  |:-------------------------------------------|:------------------------------------------------------------------------------------------------------------------|:----------------------------------|
  | <code>Ventral humeral ridge: or not</code> | <code>Metacarpals, Metacarpal I, presence: absent</code>                                                          | <code>0.05558851078197206</code>  |
  | <code>Ventral humeral ridge: or not</code> | <code>Metapterygoid–quadrate fenestra: absent</code>                                                              | <code>0.004860625129173212</code> |
  | <code>Ventral humeral ridge: or not</code> | <code>Dorsal and ventral borders of the maxillary articular process: straight or slightly curved ventrally</code> | <code>0.10380567059620477</code>  |
* Loss: [<code>CoSENTLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosentloss) with these parameters:
  ```json
  {
      "scale": 20.0,
      "similarity_fct": "pairwise_cos_sim"
  }
  ```

#### Metrics

**Semantic Similarity:**

* Datasets: `pheno-dev` and `pheno-test`
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)


### Results

| Metric              | Validation set | Test set |
|:--------------------|:----------|:-----------|
| pearson_cosine      | 0.6082    | 0.6822     |
| **spearman_cosine** | **0.625** | **0.7057** |

#### Summary

Trait2Vec embeds organismal trait descriptors in a way that preserves some of the ranking structure induced by the similarity metric of the ontology.

## Environmental Impact

Experiments were conducted using a private infrastructure, which has a carbon efficiency of 0.432 kgCO$_2$eq/kWh. A cumulative of 20 hours of computation was performed on hardware of type A100 PCIe 40/80GB (TDP of 250W).

Total emissions are estimated to be 2.16 kgCO$_2$eq of which 0 percents were directly offset.

Estimations were conducted using the [MachineLearning Impact calculator](https://mlco2.github.io/impact#compute) presented in:
```bibtex
@article{lacoste2019quantifying,
  title={Quantifying the Carbon Emissions of Machine Learning},
  author={Lacoste, Alexandre and Luccioni, Alexandra and Schmidt, Victor and Dandres, Thomas},
  journal={arXiv preprint arXiv:1910.09700},
  year={2019}
}
```

### Model Architecture and Objective

```
SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: MPNetModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 256, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
)
```
* Loss: [<code>CoSENTLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosentloss) with these parameters:
  ```json
  {
      "scale": 20.0,
      "similarity_fct": "pairwise_cos_sim"
  }
  ```

#### Software

- Python: 3.10.16
- Sentence Transformers: 3.3.1
- Transformers: 4.48.1
- PyTorch: 2.5.1.post303
- Accelerate: 1.3.0
- Datasets: 2.14.4
- Tokenizers: 0.21.0

## Citation

If you use this model in your research, please cite both it and the source model & method from which it was fine-tuned:

### Model

```bibtex
@software{trait2vec2025,
  author = {Juan Garcia and Soumyashree Kar and Jim Balhoff and Hilmar Lapp},
  doi = {10.57967/hf/6892},
  title = {Trait2Vec (Revision f39747b)},
  version = {1.0.0},
  year = {2025},
  url = {https://huggingface.co/imageomics/trait2vec}
}
```


### Source Model & Method

#### Sentence Transformers
```bibtex
@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    pages = "3982-3992",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/D19-1410/",
    doi = "10.18653/v1/D19-1410"
}
```

#### CoSENTLoss
```bibtex
@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}
```


## Acknowledgements

This work was supported by the [Imageomics Institute](https://imageomics.org), which is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under [Award #2118240](https://www.nsf.gov/awardsearch/showAward?AWD_ID=2118240) (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

## Model Card Authors

Juan Garcia

## Model Card Contact

Please open a [Discussion on the Community Tab](https://huggingface.co/imageomics/trait2vec/discussions) with any questions on the model.