DeepCoDA: personalized interpretability for compositional health data

Thomas Quinn, Dang Nguyen, Santu Rana, Sunil Gupta, Svetha Venkatesh
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:7877-7886, 2020.

Abstract

Abstract Interpretability allows the domain-expert to directly evaluate the model’s relevance and reliability, a practice that offers assurance and builds trust. In the healthcare setting, interpretable models should implicate relevant biological mechanisms independent of technical factors like data pre-processing. We define personalized interpretability as a measure of sample-specific feature attribution, and view it as a minimum requirement for a precision health model to justify its conclusions. Some health data, especially those generated by high-throughput sequencing experiments, have nuances that compromise precision health models and their interpretation. These data are compositional, meaning that each feature is conditionally dependent on all other features. We propose the Deep Compositional Data Analysis (DeepCoDA) framework to extend precision health modelling to high-dimensional compositional data, and to provide personalized interpretability through patient-specific weights. Our architecture maintains state-of-the-art performance across 25 real-world data sets, all while producing interpretations that are both personalized and fully coherent for compositional data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-quinn20a, title = {{D}eep{C}o{DA}: personalized interpretability for compositional health data}, author = {Quinn, Thomas and Nguyen, Dang and Rana, Santu and Gupta, Sunil and Venkatesh, Svetha}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {7877--7886}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/quinn20a/quinn20a.pdf}, url = {https://proceedings.mlr.press/v119/quinn20a.html}, abstract = {Abstract Interpretability allows the domain-expert to directly evaluate the model’s relevance and reliability, a practice that offers assurance and builds trust. In the healthcare setting, interpretable models should implicate relevant biological mechanisms independent of technical factors like data pre-processing. We define personalized interpretability as a measure of sample-specific feature attribution, and view it as a minimum requirement for a precision health model to justify its conclusions. Some health data, especially those generated by high-throughput sequencing experiments, have nuances that compromise precision health models and their interpretation. These data are compositional, meaning that each feature is conditionally dependent on all other features. We propose the Deep Compositional Data Analysis (DeepCoDA) framework to extend precision health modelling to high-dimensional compositional data, and to provide personalized interpretability through patient-specific weights. Our architecture maintains state-of-the-art performance across 25 real-world data sets, all while producing interpretations that are both personalized and fully coherent for compositional data.} }
Endnote
%0 Conference Paper %T DeepCoDA: personalized interpretability for compositional health data %A Thomas Quinn %A Dang Nguyen %A Santu Rana %A Sunil Gupta %A Svetha Venkatesh %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-quinn20a %I PMLR %P 7877--7886 %U https://proceedings.mlr.press/v119/quinn20a.html %V 119 %X Abstract Interpretability allows the domain-expert to directly evaluate the model’s relevance and reliability, a practice that offers assurance and builds trust. In the healthcare setting, interpretable models should implicate relevant biological mechanisms independent of technical factors like data pre-processing. We define personalized interpretability as a measure of sample-specific feature attribution, and view it as a minimum requirement for a precision health model to justify its conclusions. Some health data, especially those generated by high-throughput sequencing experiments, have nuances that compromise precision health models and their interpretation. These data are compositional, meaning that each feature is conditionally dependent on all other features. We propose the Deep Compositional Data Analysis (DeepCoDA) framework to extend precision health modelling to high-dimensional compositional data, and to provide personalized interpretability through patient-specific weights. Our architecture maintains state-of-the-art performance across 25 real-world data sets, all while producing interpretations that are both personalized and fully coherent for compositional data.
APA
Quinn, T., Nguyen, D., Rana, S., Gupta, S. & Venkatesh, S.. (2020). DeepCoDA: personalized interpretability for compositional health data. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:7877-7886 Available from https://proceedings.mlr.press/v119/quinn20a.html.

Related Material