Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets
With the increasingly more important role of machine learning (ML) models in chemical research, the need for putting a level of confidence to the model predictions naturally arises. Several methods for obtaining uncertainty estimates have been proposed in recent years but consensus on the evaluation...
| Main Authors: | , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Springer International Publishing
2024
|
| Online Access: | https://hdl.handle.net/1721.1/153303 |
| _version_ | 1811071536454434816 |
|---|---|
| author | Rasmussen, Maria H. Duan, Chenru Kulik, Heather J. Jensen, Jan H. |
| author2 | Massachusetts Institute of Technology. Department of Chemistry |
| author_facet | Massachusetts Institute of Technology. Department of Chemistry Rasmussen, Maria H. Duan, Chenru Kulik, Heather J. Jensen, Jan H. |
| author_sort | Rasmussen, Maria H. |
| collection | MIT |
| description | With the increasingly more important role of machine learning (ML) models in chemical research, the need for putting a level of confidence to the model predictions naturally arises. Several methods for obtaining uncertainty estimates have been proposed in recent years but consensus on the evaluation of these have yet to be established and different studies on uncertainties generally uses different metrics to evaluate them. We compare three of the most popular validation metrics (Spearman’s rank correlation coefficient, the negative log likelihood (NLL) and the miscalibration area) to the error-based calibration introduced by Levi et al. (Sensors 2022, 22, 5540). Importantly, metrics such as the negative log likelihood (NLL) and Spearman’s rank correlation coefficient bear little information in themselves. We therefore introduce reference values obtained through errors simulated directly from the uncertainty distribution. The different metrics target different properties and we show how to interpret them, but we generally find the best overall validation to be done based on the error-based calibration plot introduced by Levi et al. Finally, we illustrate the sensitivity of ranking-based methods (e.g. Spearman’s rank correlation coefficient) towards test set design by using the same toy model ferent test sets and obtaining vastly different metrics (0.05 vs. 0.65). |
| first_indexed | 2024-09-23T08:52:44Z |
| format | Article |
| id | mit-1721.1/153303 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T08:52:44Z |
| publishDate | 2024 |
| publisher | Springer International Publishing |
| record_format | dspace |
| spelling | mit-1721.1/1533032024-06-28T15:06:41Z Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets Rasmussen, Maria H. Duan, Chenru Kulik, Heather J. Jensen, Jan H. Massachusetts Institute of Technology. Department of Chemistry Massachusetts Institute of Technology. Department of Chemical Engineering With the increasingly more important role of machine learning (ML) models in chemical research, the need for putting a level of confidence to the model predictions naturally arises. Several methods for obtaining uncertainty estimates have been proposed in recent years but consensus on the evaluation of these have yet to be established and different studies on uncertainties generally uses different metrics to evaluate them. We compare three of the most popular validation metrics (Spearman’s rank correlation coefficient, the negative log likelihood (NLL) and the miscalibration area) to the error-based calibration introduced by Levi et al. (Sensors 2022, 22, 5540). Importantly, metrics such as the negative log likelihood (NLL) and Spearman’s rank correlation coefficient bear little information in themselves. We therefore introduce reference values obtained through errors simulated directly from the uncertainty distribution. The different metrics target different properties and we show how to interpret them, but we generally find the best overall validation to be done based on the error-based calibration plot introduced by Levi et al. Finally, we illustrate the sensitivity of ranking-based methods (e.g. Spearman’s rank correlation coefficient) towards test set design by using the same toy model ferent test sets and obtaining vastly different metrics (0.05 vs. 0.65). 2024-01-10T21:07:54Z 2024-01-10T21:07:54Z 2023-12-18 2023-12-24T04:17:48Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/153303 Journal of Cheminformatics. 2023 Dec 18;15(1):121 PUBLISHER_CC en https://doi.org/10.1186/s13321-023-00790-0 Creative Commons Attribution https://creativecommons.org/licenses/by/4.0/ The Author(s) application/pdf Springer International Publishing Springer International Publishing |
| spellingShingle | Rasmussen, Maria H. Duan, Chenru Kulik, Heather J. Jensen, Jan H. Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title | Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title_full | Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title_fullStr | Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title_full_unstemmed | Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title_short | Uncertain of uncertainties? A comparison of uncertainty quantification metrics for chemical data sets |
| title_sort | uncertain of uncertainties a comparison of uncertainty quantification metrics for chemical data sets |
| url | https://hdl.handle.net/1721.1/153303 |
| work_keys_str_mv | AT rasmussenmariah uncertainofuncertaintiesacomparisonofuncertaintyquantificationmetricsforchemicaldatasets AT duanchenru uncertainofuncertaintiesacomparisonofuncertaintyquantificationmetricsforchemicaldatasets AT kulikheatherj uncertainofuncertaintiesacomparisonofuncertaintyquantificationmetricsforchemicaldatasets AT jensenjanh uncertainofuncertaintiesacomparisonofuncertaintyquantificationmetricsforchemicaldatasets |