Hugh Chen

Hugh Chen

Auto-generated profile, based on publicly available information

Hugh Chen

Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, USA

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 May 2023 in Nature Machine Intelligence
Verified
Journal Article
Closed access
PDF file
HCHugh Chen
Hugh Chen
ICIan C. Covert
Ian C. Covert
SLScott M. Lundberg
Scott M. Lundberg
SLSu-In Lee
Su-In Lee

3 August 2022 in Nature Communications
Verified
Journal Article
Open access
PDF file
HCHugh Chen
Hugh Chen
SLScott M. Lundberg
Scott M. Lundberg
SLSu-In Lee
Su-In Lee
Local feature attribution methods are increasingly used to explain complex machine learning models. However, current methods are limited because they are extremely expensive to compute or are not capable of explaining a distributed series of models where each model is owned by a separate institution. The latter is particularly important because it often arises in finance where explanations are mandated. Here, we present Generalized DeepSHAP (G-DeepSHAP), a tractable method to propagate local feature attributions through complex series of models based on a connection to the Shapley value. We evaluate G-DeepSHAP across biological, health, and financial datasets to show that it provides equally salient explanations an order of magnitude faster than existing model-agnostic attribution techniques and demonstrate its use in an important distributed series of models setting.

1 May 2023 in Nature Biomedical Engineering
Verified
Journal Article
Closed access
PDF file
JJJoseph D. Janizek
Joseph D. Janizek
ADAyse B. Dincer
Ayse B. Dincer
SCSafiye Celik
Safiye Celik
HCHugh Chen
Hugh Chen
WCWilliam Chen
William Chen
SLSu-In Lee
Su-In Lee

8 December 2021 in npj Digital Medicine
Verified
Journal Article
Open access
PDF file
HCHugh Chen
Hugh Chen
SLScott M. Lundberg
Scott M. Lundberg
GEGabriel Erion
Gabriel Erion
JKJerry H. Kim
Jerry H. Kim
SLSu-In Lee
Su-In Lee
Hundreds of millions of surgical procedures take place annually across the world, which generate a prevalent type of electronic health record (EHR) data comprising time series physiological signals. Here, we present a transferable embedding method (i.e., a method to transform time series signals into input features for predictive machine learning models) named PHASE (PHysiologicAl Signal Embeddings) that enables us to more accurately forecast adverse surgical outcomes based on physiological signals. We evaluate PHASE on minute-by-minute EHR data of more than 50,000 surgeries from two operating room (OR) datasets and patient stays in an intensive care unit (ICU) dataset. PHASE outperforms other state-of-the-art approaches, such as long-short term memory networks trained on raw data and gradient boosted trees trained on handcrafted features, in predicting six distinct outcomes: hypoxemia, hypocapnia, hypotension, hypertension, phenylephrine, and epinephrine. In a transfer learning setting where we train embedding models in one dataset then embed signals and predict adverse events in unseen data, PHASE achieves significantly higher prediction accuracy at lower computational cost compared to conventional approaches. Finally, given the importance of understanding models in clinical applications we demonstrate that PHASE is explainable and validate our predictive models using local feature attribution methods.

6 October 2022 in medRxiv
Verified
Preprint
Open access
PDF file
WQWei Qiu
Wei Qiu
HCHugh Chen
Hugh Chen
MKMatt Kaeberlein
Matt Kaeberlein
SLSu-In Lee
Su-In Lee
Background: An individual’s biological age is a measurement of health status and provides a mechanistic understanding of aging. Age clocks estimate a biological age of an individual based on their various features . Existing clocks have key limitations caused by the undesirable tradeoff between accuracy (i.e., predictive performance for chronological age or mortality, often achieved by complex, black-box models) and interpretability (i.e., the contributions of features to biological age). Here, we present ‘ENABL ( E xplai NA ble B io L ogical) Age’, a computational framework that combines machine learning (ML) models with explainable AI (XAI) methods to accurately estimate biological age with individualized explanations. Methods: To construct ENABL Age clock, we first need to predict an age-related outcome of interest (e.g., all-cause or cause-specific mortality), and then rescale the predictions nonlinearly to estimate biological age. We trained and evaluated the ENABL Age clock using the UK Biobank (501,366 samples with 825 features) and NHANES 1999-2014 (47,084 samples with 158 features) datasets. To explain the ENABL Age clock, we extended existing XAI methods so we could linearly decompose any individual’s ENABL Age into contributing risk factors. To make ENABL Age clock broadly accessible, we developed two versions: (1) ENABL Age-L, which is based on popular blood tests, and (2) ENABL Age-Q, which is based on questionnaire features. Finally, when we created ENABL Age clocks based on predictions of different age-related outcomes, we validated that each one captures sensible, yet disparate aging mechanisms by performing GWAS association analyses. Findings: Our results indicate that ENABL Age clocks successfully separate healthy from unhealthy aging individuals and are stronger predictors of mortality than existing age clocks. We externally validated our results by training ENABL Age clocks on UK Biobank data and testing on NHANES data. The individualized explanations that reveal the contribution of specific features to ENABL Age provide insights into the important features for biological age. Association analysis with risk factors and agingrelated morbidities, and genome-wide association study (GWAS) results on ENABL Age clocks trained on different mortality causes show that each one captures sensible aging mechanisms. Interpretation: We developed and validated a new ML and XAI-based approach to calculate and interpret biological age based on multiple aging mechanisms. Our results show strong mortality prediction power, interpretability, and flexibility. ENABL Age takes a consequential step towards accurate interpretable biological age prediction built with complex, high-performance ML models. Research in context: Evidence before this study: Biological age plays an important role to understanding the mechanisms underlying aging. We search PubMed for original articles published in all languages with the terms “biological age” published until June 22, 2022. Most prior studies focus on the first generation of biological age clocks that are designed to predict chronological age. These clocks have weak and variable associations with mortality risk and other aging outcomes. Only a few studies present the second-generation of biological age clocks, which are built directly with aging outcomes. However, these studies use linear models and do not provide individualized explanations. Moreover, previous biological age clocks cannot specify what aging process they capture. Unlike our study, none of the previous studies have combined a complex machine learning (ML) model and an explainable artificial intelligence (XAI) method, which allows us to build biological ages that are both accurate and interpretable. Added value of this study: In this study, we present ENABL Age, a new approach to estimate and understand biological age that combines complex ML models and XAI method. The ENABL Age approach is designed to measure secondgeneration biological age clocks by directly predicting age-related outcomes. Our results indicate that ENABL Age accurately reflects individual health status. We also introduce two variants of ENABL Age clocks: (1) ENABL Age-L, which takes popular blood tests as inputs (usable by medical professionals), and (2) ENABL Age-Q, which takes questionnaire features as inputs (usable by non-professional healthcare consumers). We extend existing XAI methods to calculate the contributions of input features to ENABL Age estimate in units of years, which makes our biological age clocks more human-interpretable. Our association analysis and GWAS results show that ENABL Age clocks trained on different age-related outcomes can capture different aging mechanisms. Implications of all the available evidence: We develop and validate a new ML and XAI-based approach to measure and interpret biological age based on multiple aging mechanisms. Our results demonstrate that ENABL age has strong mortality prediction power, is interpretable, and is flexible. ENABL Age takes a consequential step towards applying XAI to interpret biological age models. Its flexibility allows for many future extensions to omics data, even multi-omic data, and multi-task learning.