Research map / Probabilistic and scientific modeling / Molecules and proteins
DISSOLVR: An Interpretable and Fast Framework for Aqueous and Organic Solubility Prediction
Vansh Ramani, Har Ashish Arora, Dhairya Kuchhal, Sayan Ranu, Tarak Karmakar
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 interaction terms (solute-solvent interaction layer) (ablation) | DISSOLVR Regime II · Leeds (solute holdout) | RMSE: +0.0112 RMSE when removed | Improved |
| add solute and solvent descriptors (ablation) | DISSOLVR Regime II · Leeds (solute holdout) | RMSE: 0.8764 -> 1.1001 when removed | Improved |
| add topological features (ablation) | DISSOLVR Regime II · Leeds (solute holdout) | RMSE: +0.0167 RMSE when removed | Improved |
| add physicochemical features (ablation) | DISSOLVR Regime II · Leeds (solute holdout) | RMSE: +0.0009 RMSE when removed | No clear effect |
| add physicochemical features (ablation) | DISSOLVR Regime I · AqSolDB | RMSE: 0.8110 -> 0.8950 when removed | Improved |
| add compositional, topological, energetic features (ablation) | DISSOLVR Regime I · AqSolDB | RMSE: +0.0076, +0.0101, +0.0052 when removed | Improved |
| replace cross-attention interaction block with direct concatenation of descriptors (ablation) | DISSOLVR Regime II · OOD RMSE | RMSE: triples feature dimension, about 9x training cost | Hurt |
| add hard monotonic temperature constraints (ablation) | DISSOLVR · BigSolDB, samples with conflicting data | RMSE: +0.0205 on conflicting samples vs +0.0018 on clean data | Improved |
| use explicit physical descriptors instead of latent disentanglement (RILOOD) | solvent holdout OOD · BigSolDB 1.0 | RMSE: 0.8814 -> 0.6783 | Improved |
| use DISSOLVR instead of Random Forest | Second Solubility Challenge · Second Solubility Challenge | R2: 0.5184 -> 0.4932 | Hurt |
| show LLM explanations to expert chemists | expert trust in predictions · survey of 22 chemists | trust score (1-5): 3.58 -> 3.75 | Improved |
| explain with a trained versus randomly-initialised DISSOLVR (ablation) | LLM explainer · LLM-as-judge, 30 samples | agreement (1-5): mean agreement 1.83 | Improved |
Closest papers by meaning
- SC$^3$: A Multi-Solvent Solubility Challenge and Benchmark (NeurIPS 2026 · similarity 0.88)
- CheMixHub: Datasets and Benchmarks for Chemical Mixture Property Prediction (NeurIPS 2025 · similarity 0.88)
- BOOM: Benchmarking Out-Of-distribution Molecular Property Predictions of Machine Learning Models (NeurIPS 2025 · similarity 0.86)
- Relational Invariant Learning for Robust Solvation Free Energy Prediction (ICML 2025 · similarity 0.86)
- The Catechol Benchmark: Time-series Solvent Selection Data for Few-shot Machine Learning (NeurIPS 2025 · similarity 0.85)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.85)
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.
