Just Add Data: automated predictive modeling for knowledge discovery and feature selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Translational Medicine

Background:

• Automated machine learning (AutoML) is emerging for predictive modeling.
• Omics data in translational medicine is often low-sample and high-dimensional.

Purpose of the Study:

• Introduce Just Add Data Bio (JADBio), an AutoML platform for omics data.
• Demonstrate JADBio's capability for predictive modeling and knowledge discovery.
• Highlight JADBio's focus on feature selection and biosignature identification.

Main Methods:

• Applied AutoML principles to omics data analysis.
• Utilized feature selection to identify minimal biomarker subsets (biosignatures).
• Compared JADBio against Hyper-Parameter Optimization Machine Learning libraries.

Main Results:

• JADBio identifies predictive biosignatures with a minimal number of features.
• The platform maintains competitive predictive performance.
• Accurate out-of-sample performance estimation was achieved.

Conclusions:

• JADBio offers a powerful AutoML solution for omics data in translational medicine.
• The platform facilitates clinical use through interpretable models and decision-making insights.
• JADBio effectively balances predictive accuracy with biomarker discovery.