Scholarlink

This study proposes a hybrid approach to explore metabolic flux space by integrating genome-scale metabolic models (GEMs) with deep neural networks. The method combines the mechanistic framework of the CardiacMitochondria GEM with the pattern recognition capabilities of neural networks. This allows for improved prediction of complex metabolic phenotypes. We propose and implement a deep learning-based flux (DLBF) model trained on fluxes sampled using Riemannian Hamiltonian Monte Carlo (RHMC). The model was validated using a cardiac mitochondrial dataset, achieving a coefficient of determination (R2) of 0.653, a Mean Absolute Error (MAE) of 51.67, a Mean Squared Error (MSE) of 39,477.92, and a Root Mean Squared Error (RMSE) of 20.16, confirming a moderately accurate prediction performance. Although most metabolic fluxes showed strong agreement with expected values, discrepancies in specific pathways suggest areas where biological regulation remains underrepresented. Despite using moderate computational resources, DLBF completed training within practical time limits on standard hardware. These results highlight DLBF as a predictive and interpretable tool for metabolic modeling, with potential applications in cardiovascular research, metabolic engineering, and educational settings.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning-Based Flux Prediction for Cardiac Mitochondrial Metabolism: A Comparative Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Maria Jose Chavez-Fajardo,
  • Jose Alejandro Morales Valencia,
  • German Preciat

摘要

This study proposes a hybrid approach to explore metabolic flux space by integrating genome-scale metabolic models (GEMs) with deep neural networks. The method combines the mechanistic framework of the CardiacMitochondria GEM with the pattern recognition capabilities of neural networks. This allows for improved prediction of complex metabolic phenotypes. We propose and implement a deep learning-based flux (DLBF) model trained on fluxes sampled using Riemannian Hamiltonian Monte Carlo (RHMC). The model was validated using a cardiac mitochondrial dataset, achieving a coefficient of determination (R2) of 0.653, a Mean Absolute Error (MAE) of 51.67, a Mean Squared Error (MSE) of 39,477.92, and a Root Mean Squared Error (RMSE) of 20.16, confirming a moderately accurate prediction performance. Although most metabolic fluxes showed strong agreement with expected values, discrepancies in specific pathways suggest areas where biological regulation remains underrepresented. Despite using moderate computational resources, DLBF completed training within practical time limits on standard hardware. These results highlight DLBF as a predictive and interpretable tool for metabolic modeling, with potential applications in cardiovascular research, metabolic engineering, and educational settings.