Evidential Deep Learning for Guided Molecular Property Prediction and Discovery
While neural networks achieve state-of-the-art performance for many molecular modeling and structure-property prediction tasks, these models can struggle with generalization to out-of-domain examples, exhibit poor sample efficiency, and produce uncalibrated predictions. In this paper, we leverage ad...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
American Chemical Society (ACS)
2022
|
| Online Access: | https://hdl.handle.net/1721.1/142794 |
| _version_ | 1811069639486078976 |
|---|---|
| author | Soleimany, Ava P Amini, Alexander Goldman, Samuel Rus, Daniela Bhatia, Sangeeta N Coley, Connor W |
| author2 | Massachusetts Institute of Technology. Department of Chemical Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Chemical Engineering Soleimany, Ava P Amini, Alexander Goldman, Samuel Rus, Daniela Bhatia, Sangeeta N Coley, Connor W |
| author_sort | Soleimany, Ava P |
| collection | MIT |
| description | While neural networks achieve state-of-the-art performance for many molecular modeling and structure-property prediction tasks, these models can struggle with generalization to out-of-domain examples, exhibit poor sample efficiency, and produce uncalibrated predictions. In this paper, we leverage advances in evidential deep learning to demonstrate a new approach to uncertainty quantification for neural network-based molecular structure-property prediction at no additional computational cost. We develop both evidential 2D message passing neural networks and evidential 3D atomistic neural networks and apply these networks across a range of different tasks. We demonstrate that evidential uncertainties enable (1) calibrated predictions where uncertainty correlates with error, (2) sample-efficient training through uncertainty-guided active learning, and (3) improved experimental validation rates in a retrospective virtual screening campaign. Our results suggest that evidential deep learning can provide an efficient means of uncertainty quantification useful for molecular property prediction, discovery, and design tasks in the chemical and physical sciences. |
| first_indexed | 2024-09-23T08:13:34Z |
| format | Article |
| id | mit-1721.1/142794 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T08:13:34Z |
| publishDate | 2022 |
| publisher | American Chemical Society (ACS) |
| record_format | dspace |
| spelling | mit-1721.1/1427942023-04-18T15:19:44Z Evidential Deep Learning for Guided Molecular Property Prediction and Discovery Soleimany, Ava P Amini, Alexander Goldman, Samuel Rus, Daniela Bhatia, Sangeeta N Coley, Connor W Massachusetts Institute of Technology. Department of Chemical Engineering Harvard University--MIT Division of Health Sciences and Technology Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology. Computational and Systems Biology Program Howard Hughes Medical Institute While neural networks achieve state-of-the-art performance for many molecular modeling and structure-property prediction tasks, these models can struggle with generalization to out-of-domain examples, exhibit poor sample efficiency, and produce uncalibrated predictions. In this paper, we leverage advances in evidential deep learning to demonstrate a new approach to uncertainty quantification for neural network-based molecular structure-property prediction at no additional computational cost. We develop both evidential 2D message passing neural networks and evidential 3D atomistic neural networks and apply these networks across a range of different tasks. We demonstrate that evidential uncertainties enable (1) calibrated predictions where uncertainty correlates with error, (2) sample-efficient training through uncertainty-guided active learning, and (3) improved experimental validation rates in a retrospective virtual screening campaign. Our results suggest that evidential deep learning can provide an efficient means of uncertainty quantification useful for molecular property prediction, discovery, and design tasks in the chemical and physical sciences. 2022-05-27T15:04:36Z 2022-05-27T15:04:36Z 2021 2022-05-27T14:46:22Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/142794 Soleimany, Ava P, Amini, Alexander, Goldman, Samuel, Rus, Daniela, Bhatia, Sangeeta N et al. 2021. "Evidential Deep Learning for Guided Molecular Property Prediction and Discovery." ACS Central Science, 7 (8). en 10.1021/ACSCENTSCI.1C00546 ACS Central Science Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Licens http://creativecommons.org/licenses/by-nc-nd/4.0/ application/pdf American Chemical Society (ACS) American Chemical Society |
| spellingShingle | Soleimany, Ava P Amini, Alexander Goldman, Samuel Rus, Daniela Bhatia, Sangeeta N Coley, Connor W Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title | Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title_full | Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title_fullStr | Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title_full_unstemmed | Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title_short | Evidential Deep Learning for Guided Molecular Property Prediction and Discovery |
| title_sort | evidential deep learning for guided molecular property prediction and discovery |
| url | https://hdl.handle.net/1721.1/142794 |
| work_keys_str_mv | AT soleimanyavap evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery AT aminialexander evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery AT goldmansamuel evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery AT rusdaniela evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery AT bhatiasangeetan evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery AT coleyconnorw evidentialdeeplearningforguidedmolecularpropertypredictionanddiscovery |