A comprehensive multi-task deep learning approach for predicting metabolic syndrome with genetic, nutritional, and clinical data

Abstract: Metabolic syndrome (MetS) is a complex condition characterized by a group of interconnected metabolic abnormalities. Due to its increasing prevalence, better predictive markers are needed. Therefore, this study aims to develop predictive models for MetS by integrating adipokines, metabolic and cardiovascular risk factors, and anthropometric indices. Data were collected from 381 subjects aged 20 to 59 years (242 women and 139 men) from Guadalajara, Jalisco, Mexico, who were classified as having MetS or non-MetS based on the ATP-III criteria. Four supervised machine learning models were developed—Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)—and their performance was evaluated using the Area under the Curve (AUC), calibration curves, Decision Curve Analysis (DCA), and local interpretability analysis. The RF and XGBoost models achieved the highest AUCs (0.940 and 0.954). The RF and LR models were the best calibrated and showed the highest net benefit in DCA. Key variables included age, anthropometric indices (BRI and DAI), insulin resistance measures (HOMA-IR), lipid profiles (sdLDL-C and LDL-C), and high-molecular-weight adiponectin, used to classify the presence of MetS. The results highlight the usefulness of specific models and the importance of anthropometric variables, cardiovascular risk factors, metabolic profiles, and adiponectin as indicators of MetS.

TL;DR: A general gradient descent boosting paradigm is developed for additive expansions based on any fitting criterion, and specific algorithms are presented for least-squares, least absolute deviation, and Huber-M loss functions for regression, and multiclass logistic likelihood for classification.

TL;DR: In this article, the authors propose LIME, a method to explain models by presenting representative individual predictions and their explanations in a non-redundant way, framing the task as a submodular optimization problem.

TL;DR: It is proved that, since the data instances with larger gradients play a more important role in the computation of information gain, GOSS can obtain quite accurate estimation of the information gain with a much smaller data size, and is called LightGBM.

TL;DR: This article seeks to help ML practitioners apply MTL by shedding light on how MTL works and providing guidelines for choosing appropriate auxiliary tasks, particularly in deep neural networks.