Calculation of exact Shapley values for support vector machines with Tanimoto kernel enables model interpretation

Feldmann, Christian; Bajorath, Jürgen

doi:10.1016/j.isci.2022.105023

Cited by 13 publications

(15 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We have set out to derive multilayer explainable machine learning (XML) models − for predicting CB 1 R and CB 2 R ligands. Specifically, four cooperating classification models were generated for predicting and rationalizing molecular determinants driving selective ligand binding to CB 1 R and CB 2 R. Each classification model was independently derived based on different sets of training data carefully curated from the ChEMBL database (release 31) .…”

Section: Introductionmentioning

confidence: 99%

CIRCE: Web-Based Platform for the Prediction of Cannabinoid Receptor Ligands Using Explainable Machine Learning

Gambacorta,

Ciriaco,

Amoroso

et al. 2023

J. Chem. Inf. Model.

Self Cite

View full text Add to dashboard Cite

The endocannabinoid system, which includes cannabinoid receptor 1 and 2 subtypes (CB 1 R and CB 2 R, respectively), is responsible for the onset of various pathologies including neurodegeneration, cancer, neuropathic and inflammatory pain, obesity, and inflammatory bowel disease. Given the high similarity of CB 1 R and CB 2 R, generating subtype-selective ligands is still an open challenge. In this work, the Cannabinoid Iterative Revaluation for Classification and Explanation (CIRCE) compound prediction platform has been generated based on explainable machine learning to support the design of selective CB 1 R and CB 2 R ligands. Multilayer classifiers were combined with Shapley value analysis to facilitate explainable predictions. In test calculations, CIRCE predictions reached ∼80% accuracy and structural features determining ligand predictions were rationalized. CIRCE was designed as a web-based prediction platform that is made freely available as a part of our study.

show abstract

Section: Introductionmentioning

confidence: 99%

CIRCE: Web-Based Platform for the Prediction of Cannabinoid Receptor Ligands Using Explainable Machine Learning

Gambacorta,

Ciriaco,

Amoroso

et al. 2023

J. Chem. Inf. Model.

Self Cite

View full text Add to dashboard Cite

show abstract

“…For SVM, the use of the Tanimoto kernel was mandatory to enable the calculation of exact Shapley values (which is currently not possible for other kernels) 29 . Approximated SVM Shapley values only poorly correlated with exact values 29 , which was insufficient for accurate model explanation. For comparison, SVM compound classification was repeated with an alternative (RBF) kernel, yielding nearly indistinguishable prediction accuracy compared to the Tanimoto kernel (Supplementary Fig.…”

Section: Resultsmentioning

confidence: 99%

“…Therefore, we carried out a comparative Shapley value analysis (see "Methods") across all models. For decision tree methods such as RF and SVM employing the Tanimoto kernel, exact Shapley values can be calculated using the TreeExplainer 28 and Shapley Value-Expressed Tanimoto Similarity (SVETA) 29 methods, respectively (instead of www.nature.com/scientificreports/ locally approximated values as for other ML methods using SHAP 16 ). For SVM, the use of the Tanimoto kernel was mandatory to enable the calculation of exact Shapley values (which is currently not possible for other kernels) 29 .…”

Section: Rationalizing Predictionsmentioning

confidence: 99%

“…For decision tree methods such as RF and SVM employing the Tanimoto kernel, exact Shapley values can be calculated using the TreeExplainer 28 and Shapley Value-Expressed Tanimoto Similarity (SVETA) 29 methods, respectively (instead of www.nature.com/scientificreports/ locally approximated values as for other ML methods using SHAP 16 ). For SVM, the use of the Tanimoto kernel was mandatory to enable the calculation of exact Shapley values (which is currently not possible for other kernels) 29 . Approximated SVM Shapley values only poorly correlated with exact values 29 , which was insufficient for accurate model explanation.…”

Section: Rationalizing Predictionsmentioning

confidence: 99%

See 1 more Smart Citation

Differences in learning characteristics between support vector machine and random forest models for compound classification revealed by Shapley value analysis

Siemers

Bajorath

2023

Sci Rep

Self Cite

View full text Add to dashboard Cite

The random forest (RF) and support vector machine (SVM) methods are mainstays in molecular machine learning (ML) and compound property prediction. We have explored in detail how binary classification models derived using these algorithms arrive at their predictions. To these ends, approaches from explainable artificial intelligence (XAI) are applicable such as the Shapley value concept originating from game theory that we adapted and further extended for our analysis. In large-scale activity-based compound classification using models derived from training sets of increasing size, RF and SVM with the Tanimoto kernel produced very similar predictions that could hardly be distinguished. However, Shapley value analysis revealed that their learning characteristics systematically differed and that chemically intuitive explanations of accurate RF and SVM predictions had different origins.

show abstract

“…The popularity of GNNs has also been accompanied by an increasing need for explainability, [5][6][7][8][9][10][11][12][13] as these models have been notoriously known for their black-box character. Towards this goal, explainable artificial intelligence techniques, such as feature attribution analyses, have become relevant tools.…”

Section: Introductionmentioning

confidence: 99%

Explaining compound activity predictions with a substructure-aware loss for graph neural networks

Amara

Rodríguez-Pérez

Jiménez-Luna

2023

Preprint

View full text Add to dashboard Cite

Explainable machine learning is increasingly used in drug discovery to help rationalize compound property predictions. Feature attribution techniques are popular choices to identify which molecular substructures are responsible for a predicted property change. However, established molecular feature attribution methods have so far displayed low performance for popular deep learning algorithms such as graph neural networks (GNNs), especially when compared with simpler modeling alternatives such as random forests coupled with atom masking. To mitigate this problem, in this work a modification of the regression objective for GNNs is proposed to specifically account for common core structures between pairs of molecules. The presented approach showed higher accuracy on a recently-proposed explainability benchmark. This methodology has the potential to assist with model explainability in drug discovery pipelines, particularly in lead optimization efforts where specific chemical series are investigated.

show abstract

Calculation of exact Shapley values for support vector machines with Tanimoto kernel enables model interpretation

Cited by 13 publications

References 36 publications

CIRCE: Web-Based Platform for the Prediction of Cannabinoid Receptor Ligands Using Explainable Machine Learning

CIRCE: Web-Based Platform for the Prediction of Cannabinoid Receptor Ligands Using Explainable Machine Learning

Differences in learning characteristics between support vector machine and random forest models for compound classification revealed by Shapley value analysis

Explaining compound activity predictions with a substructure-aware loss for graph neural networks

Contact Info

Product

Resources

About