Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 56 min ago
PreprintCC BYChemRxiv

Molecular Property Prediction with Well-Calibrated Uncertainties via Graph Neural Networks and Conformal Quantile Regression

  • Shih-Cheng Li1,2
  • Charlles R. A. Abreu1,2
  • Markus Kraft3,4
  • William H. Green1,2
4 affiliations
  1. 1Department of Chemical Engineering
  2. 2Massachusetts Institute of Technology
  3. 3Department of Chemical Engineering and Biotechnology
  4. 4University of Cambridge
Open at sourcePDFDOICite

Abstract

Accurate predictive uncertainty quantification (pUQ) is critical for assessing chemical property predictions. Incorrect quantification could lead to costly synthesis and characterization of drug candidates destined to fail, or result in misguided environmental regulations. This work integrates Conformal Quantile Regression (CQR) with graph neural networks to achieve accurate molecular property prediction and well-calibrated predictive uncertainty quantification. By combining conformal prediction with quantile regression, CQR provides model-agnostic, distribution-free prediction intervals with valid coverage for finite samples. We benchmark it against established pUQ approaches, including ensembling, Monte Carlo dropout sampling, mean-variance estimation, and evidential regression. Our experiments on datasets spanning quantum mechanics, physical chemistry, and pharmacokinetics show that CQR generally provides strong calibration while maintaining competitive predictive accuracy and uncertainty-ranking performance. It also effectively captures heteroskedastic noise across the data, which is a key requirement for robust pUQ. Furthermore, we demonstrate its efficacy with an uncertainty-guided retrospective virtual screening experiment for antibiotic discovery, achieving hit rates above 90%. These results establish CQR as a promising, scalable tool for pUQ in molecular property prediction, supporting informed decision-making in applications such as active learning and compound screening for drug discovery and materials design.

Related

  1. Prediction of Singlet-Triplet Gaps Using Self-Attention–Based Regression with Model Integration into a Molecular Editor
    Péter Pál Fehér, András Stirling, Andras Stirling
    ChemRxivPreprintCC BY-NCFull textMaterials & polymersChemRxiv
  2. Physics-Guided Learning with Calibrated Uncertainty for Molecular Thermodynamics Beyond Observed Chemical Space
    Yi-Cheng Shen, Shu Wang, Shiang-Tai Lin
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
  3. Accept or abstain? A pre-specified test of reliability gates calibrated on held-out chemistry for molecular property and reaction-yield models
    Oleksandr Gurbych
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
  4. COSMO-RS prediction of hydrocarbons EACNs calibrated on the n-alkanes reference scale
    Jean-Marie Aubry, Théophile Gaudin, Andreas Klamt
    Journal of Molecular LiquidsPeer-reviewedCC BY 4.0Full textTheoretical & computational chemistryHAL
  5. Conformal Prediction and Multi-Ensemble Gradient Boosting for Subtype-Selective Binding Affinity Estimation Across Human Adenosine Receptors
    Utkarsh Patel
    ChemRxivPreprintCC BYFull textGeneral chemistryChemRxiv

Cite

BibTeXRIS
BibTeX
@misc{li2026molecular,
  title = {{Molecular Property Prediction with Well-Calibrated Uncertainties via Graph Neural Networks and Conformal Quantile Regression}},
  author = {Shih-Cheng Li and Charlles R. A. Abreu and Markus Kraft and William H. Green},
  howpublished = {ChemRxiv},
  year = {2026},
  month = oct,
  doi = {10.26434/chemrxiv.15010324/v1},
  url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010324/v1},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - Molecular Property Prediction with Well-Calibrated Uncertainties via Graph Neural Networks and Conformal Quantile Regression
AU  - Shih-Cheng Li
AU  - Charlles R. A. Abreu
AU  - Markus Kraft
AU  - William H. Green
PY  - 2026
DA  - 2026/10/12
PB  - ChemRxiv
DO  - 10.26434/chemrxiv.15010324/v1
UR  - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010324/v1
AB  - Accurate predictive uncertainty quantification (pUQ) is critical for assessing chemical property predictions. Incorrect quantification could lead to costly synthesis and characterization of drug candidates destined to fail, or result in misguided environmental regulations. This work integrates Conformal Quantile Regression (CQR) with graph neural networks to achieve accurate molecular property prediction and well-calibrated predictive uncertainty quantification. By combining conformal prediction with quantile regression, CQR provides model-agnostic, distribution-free prediction intervals with valid coverage for finite samples. We benchmark it against established pUQ approaches, including ensembling, Monte Carlo dropout sampling, mean-variance estimation, and evidential regression. Our experiments on datasets spanning quantum mechanics, physical chemistry, and pharmacokinetics show that CQR generally provides strong calibration while maintaining competitive predictive accuracy and uncertainty-ranking performance. It also effectively captures heteroskedastic noise across the data, which is a key requirement for robust pUQ. Furthermore, we demonstrate its efficacy with an uncertainty-guided retrospective virtual screening experiment for antibiotic discovery, achieving hit rates above 90%. These results establish CQR as a promising, scalable tool for pUQ in molecular property prediction, supporting informed decision-making in applications such as active learning and compound screening for drug discovery and materials design.
ER  -