Automated machine learning for endemic active tuberculosis prediction from multiplex serological data.

Area of Science:

• Immunology
• Computational Biology
• Medical Diagnostics

Background:

• Serological diagnosis of active tuberculosis (TB) relies on detecting multiple antibodies due to varied patient immune responses.
• Interpreting complex serological data for TB diagnosis requires sophisticated algorithms, which are often time-consuming to develop.

Purpose of the Study:

• To introduce and evaluate the Machine Intelligence Learning Optimizer (MILO) platform for automated generation and optimization of serological diagnostic models for TB.
• To identify the most robust antigen panel for TB serological diagnosis using machine learning.

Main Methods:

• Utilized the MILO platform for automated data processing, feature selection, model training, and validation.
• Generated and evaluated thousands of models using a 31-antigen panel.
• Tested model generalizability on secondary (TB vs. healthy) and tertiary (TB vs. COPD) out-of-sample datasets.

Main Results:

• A 23-antigen model demonstrated superior robustness across datasets.
• Achieved 90.5% sensitivity for TB detection.
• Reported specificities of 100.0% (vs. healthy) and 74.6% (vs. COPD).

Conclusions:

• MILO provides an efficient, end-to-end solution for developing optimized diagnostic models.
• The developed 23-antigen model shows significant potential for accurate TB serological diagnosis.
• MILO is suitable for rapid clinical implementation, particularly for emerging infectious diseases.