Predictive and Explanatory Uncertainties in Graph Neural Networks: A Case Study in Molecular Property Prediction

Marisa Wodrich, Aasa Feragen, Mikkel N. Schmidt
Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL), PMLR 307:487-495, 2026.

Abstract

Accurate molecular property prediction is a key challenge in fields such as drug discovery and materials science, where deep learning models offer promising solutions. However, the widespread use of these models is hindered by their lack of transparency and the difficulty in assessing the reliability of their predictions. In this study, we address these issues by integrating uncertainty quantification and explainable AI techniques to enhance the trustworthiness of graph neural networks for molecular property prediction. We focus on predicting two distinct properties: aqueous solubility and mutagenicity. By deriving explanations in the form of substructure attribution scores, we obtain interpretable explanations that signify which chemically meaningful substructures influence the model’s predictions. We incorporate uncertainty quantification to evaluate the confidence of both the predictions and their explanations. Our results demonstrate that predictive uncertainty scores correlate with the accuracy of the predictions for both tasks. Uncertainties in the explanations also correlate with prediction correctness, and there is a weak to moderate correlation between the uncertainties in the predictions and those in the explanations. These findings highlight the potential of combining uncertainty quantification and explainability to improve the trustworthiness of molecular property prediction models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v307-wodrich26a, title = {Predictive and Explanatory Uncertainties in Graph Neural Networks: A Case Study in Molecular Property Prediction}, author = {Wodrich, Marisa and Feragen, Aasa and Schmidt, Mikkel N.}, booktitle = {Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL)}, pages = {487--495}, year = {2026}, editor = {Kim, Hyeongji and Ramírez Rivera, Adín and Ricaud, Benjamin}, volume = {307}, series = {Proceedings of Machine Learning Research}, month = {06--08 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v307/main/assets/wodrich26a/wodrich26a.pdf}, url = {https://proceedings.mlr.press/v307/wodrich26a.html}, abstract = {Accurate molecular property prediction is a key challenge in fields such as drug discovery and materials science, where deep learning models offer promising solutions. However, the widespread use of these models is hindered by their lack of transparency and the difficulty in assessing the reliability of their predictions. In this study, we address these issues by integrating uncertainty quantification and explainable AI techniques to enhance the trustworthiness of graph neural networks for molecular property prediction. We focus on predicting two distinct properties: aqueous solubility and mutagenicity. By deriving explanations in the form of substructure attribution scores, we obtain interpretable explanations that signify which chemically meaningful substructures influence the model’s predictions. We incorporate uncertainty quantification to evaluate the confidence of both the predictions and their explanations. Our results demonstrate that predictive uncertainty scores correlate with the accuracy of the predictions for both tasks. Uncertainties in the explanations also correlate with prediction correctness, and there is a weak to moderate correlation between the uncertainties in the predictions and those in the explanations. These findings highlight the potential of combining uncertainty quantification and explainability to improve the trustworthiness of molecular property prediction models.} }
Endnote
%0 Conference Paper %T Predictive and Explanatory Uncertainties in Graph Neural Networks: A Case Study in Molecular Property Prediction %A Marisa Wodrich %A Aasa Feragen %A Mikkel N. Schmidt %B Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL) %C Proceedings of Machine Learning Research %D 2026 %E Hyeongji Kim %E Adín Ramírez Rivera %E Benjamin Ricaud %F pmlr-v307-wodrich26a %I PMLR %P 487--495 %U https://proceedings.mlr.press/v307/wodrich26a.html %V 307 %X Accurate molecular property prediction is a key challenge in fields such as drug discovery and materials science, where deep learning models offer promising solutions. However, the widespread use of these models is hindered by their lack of transparency and the difficulty in assessing the reliability of their predictions. In this study, we address these issues by integrating uncertainty quantification and explainable AI techniques to enhance the trustworthiness of graph neural networks for molecular property prediction. We focus on predicting two distinct properties: aqueous solubility and mutagenicity. By deriving explanations in the form of substructure attribution scores, we obtain interpretable explanations that signify which chemically meaningful substructures influence the model’s predictions. We incorporate uncertainty quantification to evaluate the confidence of both the predictions and their explanations. Our results demonstrate that predictive uncertainty scores correlate with the accuracy of the predictions for both tasks. Uncertainties in the explanations also correlate with prediction correctness, and there is a weak to moderate correlation between the uncertainties in the predictions and those in the explanations. These findings highlight the potential of combining uncertainty quantification and explainability to improve the trustworthiness of molecular property prediction models.
APA
Wodrich, M., Feragen, A. & Schmidt, M.N.. (2026). Predictive and Explanatory Uncertainties in Graph Neural Networks: A Case Study in Molecular Property Prediction. Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL), in Proceedings of Machine Learning Research 307:487-495 Available from https://proceedings.mlr.press/v307/wodrich26a.html.

Related Material