|
|
--- |
|
|
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. |
