A united model for diagnosing pulmonary tuberculosis with random forest and artificial neural network

Area of Science:

• Genomics
• Bioinformatics
• Infectious Diseases

Background:

• Pulmonary tuberculosis (PTB) is a prevalent chronic infectious disease.
• Sputum smear tests, the current gold standard for PTB diagnosis, have significant limitations in sensitivity, specificity, and sample adequacy.

Purpose of the Study:

• To identify specific genetic biomarkers for PTB.
• To construct and validate a diagnostic model for PTB using machine learning algorithms.

Main Methods:

• Utilized two public Gene Expression Omnibus (GEO) cohorts (GSE83456 for training, GSE42834 for validation).
• Employed Random Forest (RF) for biomarker identification and Artificial Neural Network (ANN) for model construction.
• Validated model accuracy using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC).
• Analyzed immunocyte proportions using the CIBERSORT algorithm.

Main Results:

• Identified 11 specific PTB biomarkers through RF classification.
• The ANN model achieved an AUC of 1.000 in the training cohort and 0.946 in the validation cohort.
• Demonstrated outstanding diagnostic performance for PTB using the developed model.

Conclusions:

• Successfully identified genetic biomarkers and developed a highly accurate blood-based diagnostic model for PTB.
• The model offers a reliable tool for early PTB detection.
• Provides new insights into PTB pathogenesis.