Research map / Architectures and efficiency / Pruning and efficient tuning
Low rank adaptation of chemical foundation models generate effective odorant representations
Grant D. McConachie, Emily Duniec, Florence V. Guerina, Meg Younger, Brian DePasquale
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add LoRA fine-tuning of the chemical model (LORAX vs frozen ProSmith) | ProSmith with ChemBERTa-77M-MTR · Carey (random 5-fold) | R2: 0.703 -> 0.712 | No clear effect |
| add LoRA fine-tuning of the chemical model | ProSmith · Carey (unseen odorants) | R2: -0.402 -> 0.045 | Improved |
| add LoRA fine-tuning of the chemical model | ProSmith · Carey (unseen receptors) | R2: 0.088 -> 0.060 | No clear effect |
| add LoRA fine-tuning of the chemical model | ProSmith · M2OR (full, weighted) | MCC: 0.641 -> 0.651 | Improved |
| add LoRA fine-tuning of the chemical model | ProSmith · M2OR (full, weighted) | AUROC: 90.47 -> 83.24 | Hurt |
| add LoRA fine-tuning of the protein model on top of chemical LoRA (ablation) | LORAX (C) · M2OR (full, weighted) | MCC: 0.651 -> 0.650 | Hurt |
| add receptor (protein) representation to the odorant-only model | ridge model with ChemBERTa-77M-MTR · Carey (feature-based, 5-fold) | R2: 0.218 -> 0.290 | Improved |
| add transformer and XGBoost ensemble (ProSmith) over ridge MP | ChemBERTa-77M-MTR features · Carey (feature-based, 5-fold) | R2: 0.290 -> 0.704 | Improved |
| replace random odorant vectors with a pretrained chemical foundation model | ProSmith · Carey (feature-based, 5-fold) | R2: 0.586 -> 0.704 | Improved |
| replace physicochemical descriptors (RDKit) with ChemBERTa foundation features | ProSmith · Carey (feature-based, 5-fold) | R2: 0.720 -> 0.704 | No clear effect |
| replace random vectors with ChemBERTa features in the ridge MP model | MP · Carey (feature-based, 5-fold) | R2: 0.294 -> 0.290 | No clear effect |
Closest papers by meaning
- Can Transformers Smell Like Humans? (NeurIPS 2024 · similarity 0.87)
- Inference with correlated priors using sisters cells (NeurIPS 2025 · similarity 0.83)
- TinyChemVL: Advancing Chemical Vision-Language Models via Efficient Visual Token Reduction and Complex Reaction Tasks (AAAI 2026 · similarity 0.81)
- Seemingly Redundant Modules Enhance Robust Odor Learning in Fruit Flies (NeurIPS 2025 · similarity 0.81)
- MolTailor: Tailoring Chemical Molecular Representation to Specific Tasks via Text Prompts (AAAI 2024 · similarity 0.80)
- Poisson Variational Autoencoder (NeurIPS 2024 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
