Downloads · 30 days
0
molecular-odor-prediction
batterdaysahead/molecular-odor-prediction
molecular-odor-prediction is a tabular classification model from batterdaysahead. Use it for the tabular classification task on the model card, and read the license before you ship it in a product. It is set up for sklearn. The card lists the license as mit.
Predict odor descriptors and perceptual ratings from a molecule's SMILES string.
Access
Public
Updated Mar 4, 2026
Repo size
11.1 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.json33 MB · 75%
At a glance
- Task
- Tabular Classification
- Library
- sklearn
- License
- mit
- Access
- Public
- Created
- Mar 4, 2026
- Updated
- Mar 4, 2026
- SHA
- 061ba2b9
Model card
From the Hugging Face model README
Odor Prediction Model
Predict odor descriptors and perceptual ratings from a molecule's SMILES string.
What it does:
- Predicts 112 odor descriptors (fruity, floral, woody, sweet, etc.)
- Predicts 3 perceptual ratings (Pleasantness, Intensity, Familiarity)
Results
| Task | Metric | Score |
|---|---|---|
| Odor Classification | Top-20 Macro F1 | 0.45 |
| Odor Classification | Mean AUC-ROC | 0.84 |
| Pleasantness | Pearson r | 0.55 |
| Intensity | Pearson r | 0.34 |
| Familiarity | Pearson r | 0.44 |
Quick Start
pip install torch xgboost rdkit numpy
from src.predict import predict
predict("c1ccc(cc1)C=O") # Benzaldehyde - smells like almonds
Output:
Odor descriptors:
sweet : 0.98
spicy : 0.98
almond : 0.98
floral : 0.96
green : 0.93
Perceptual ratings (0-1 scale):
Pleasantness: 0.99
Intensity : 0.89
Familiarity : 0.92
Data Sources
All training data comes from the Pyrfume project:
| Dataset | What it contains |
|---|---|
| GoodScents | ~4,500 molecules with expert odor annotations |
| Leffingwell | ~3,200 molecules with odor descriptors |
| Keller et al. 2016/2012 | Perceptual ratings (pleasantness, intensity, familiarity) |
| MA et al. 2021 | Additional perceptual ratings |
| Arshamian et al. 2022 | Pleasantness rankings |
Model Details
Classifier (odor descriptors):
- XGBoost with 300 trees
- Morgan fingerprints (2048-bit) from RDKit
- Per-label threshold tuning
Regressor (perceptual ratings):
- PyTorch MLP (1024-512-256 hidden layers)
- Separate models for each rating
Limitations
- Only works on single molecules, not mixtures
- Intensity predictions are weaker (r=0.34)
- Perceptual ratings trained on ~500 molecules only
Citation
@misc{molecular-odor-prediction,
title={Molecular Odor Prediction},
author={batterdaysahead},
year={2026},
url={https://huggingface.co/batterdaysahead/molecular-odor-prediction},
}
License
MIT
- Task
- Tabular Classification
- Library
- sklearn
- License
- mit
- Created
- Mar 4, 2026
- Updated
- Mar 4, 2026