Compare the performance of multiple binary classification models in microbial high-throughput sequencing datasets.

Area of Science:

• Microbial Ecology
• Bioinformatics
• Computational Biology

Background:

• Machine learning and deep learning offer tools for predicting microbial community responses to environmental shifts using high-throughput sequencing data.
• Limited research has compared the performance and practicality of various binary classification models for analyzing complex microbiota datasets.

Purpose of the Study:

• To evaluate and compare the performance, accuracy, and running time of four binary classification models: random forest (RF), support vector machine (SVM), logistic regression (LR), and back propagation neural network (BPNN).
• To identify the most suitable algorithms for microbiota data analysis and provide a framework for applying artificial intelligence in microbial ecology.

Main Methods:

• Pre-processing of microbiota datasets, including removal of low-quality variables and addressing class imbalance.
• Development and optimization of RF, SVM, LR, and BPNN models.
• Tuning of model parameters, specifically epochs for BPNN and n_estimators for RF, through iterative adjustments.

Main Results:

• Dataset pre-processing is crucial for effective model construction.
• BPNN and RF emerged as the most suitable methods for building microbiota binary classification models.
• BPNN exhibited the highest accuracy and most robust performance, followed closely by RF, across multiple microbial datasets.

Conclusions:

• The study provides a comparative analysis of machine learning and deep learning models for microbiota data.
• BPNN and RF are recommended for predicting microbial responses, with BPNN offering superior accuracy.
• The findings offer a roadmap for utilizing artificial intelligence in microbial ecology research.