Optimizing intervertebral disc cell metabolic phenotyping with machine learning and artificial neural networks

Area of Science:

• Cellular Metabolism and Bioenergetics
• Computational Biology and Bioinformatics
• Biomedical Research and Drug Discovery

Background:

• Cellular metabolic phenotyping is crucial for understanding health and disease.
• Seahorse XF analyzers measure oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) for insights into mitochondrial and glycolytic activity.
• Accurate phenotyping demands meticulous experimental preparation and reagent titration.

Purpose of the Study:

• To optimize Seahorse bioenergetic flux analysis for in vitro metabolic phenotyping using machine learning (ML) and artificial neural networks (ANN).
• To enhance the accuracy of metabolic data prediction and reduce the need for extensive experimental testing.
• To develop a data-driven framework for high-resolution metabolic phenotyping.

Main Methods:

• Applied supervised ML techniques (Regression and Classification Learner App) to evaluate predictive capabilities of various algorithms.
• Assessed ANN and ML model performance using metrics like root mean square error (RMSE), mean square error (MSE), mean absolute error, and accuracy percentage.
• Integrated unsupervised ML with a supervised ANN to create a comprehensive biological workflow.

Main Results:

• Fine and boosted trees in regression models, and fine/medium trees and linear/quadratic support vector machines (SVM) in classification models, significantly enhanced prediction accuracy.
• The integrated ML and ANN approach closely aligned with experimental data, minimizing discrepancies between predicted and observed values.
• Achieved a robust, unbiased, and data-driven framework for metabolic phenotyping.

Conclusions:

• The combination of precise experimental biology and advanced computational analysis, specifically ML and ANN, provides a powerful synergy for metabolic phenotyping.
• This approach establishes a rigorous workflow for bioenergetic phenotyping, improving accuracy and efficiency.
• The framework holds significant promise for advancing fundamental research and accelerating therapeutic discovery.