Toward Multilabel Classification for Multiple Disease Prediction Using Gut Microbiota Profiles.

Area of Science:

• Microbiome Research
• Computational Biology
• Human Health

Background:

• High-throughput sequencing provides extensive human gut microbiome data, driving research into its links with complex diseases.
• Machine learning (ML) aids in identifying individuals at high risk for diseases based on microbiome profiles.
• Existing ML methods face challenges with microbial data heterogeneity, sparsity, and understanding inter-disease relationships.

Purpose of the Study:

• To develop an innovative multilabel classification (MLC) framework, GutMLC, for human gut microbiome-based disease detection.
• To integrate prior knowledge of entity semantic similarity to improve feature selection and loss functions in MLC.
• To address label imbalance issues within and between disease labels using an adapted focal loss (FL) function.

Main Methods:

• Reframing disease detection as a multilabel classification (MLC) problem.
• Incorporating entity semantic similarity into multilabel feature selection and loss functions.
• Adapting the focal loss (FL) function for MLC with debiased inverse weighting to handle label imbalance.

Main Results:

• GutMLC demonstrated competitive performance against common MLC and single-label classification (SLC) algorithms in extensive experiments.
• The framework effectively captures shared attributes and inherent associations among diseases and microbes.
• The approach shows promise in handling the heterogeneity and sparsity of microbial features.

Conclusions:

• GutMLC offers a robust framework for microbiome-based multilabel disease prediction.
• The study highlights the potential of gut microbiota as reliable biomarkers for predicting multiple diseases simultaneously.
• Integrating semantic similarity and addressing label imbalance are key advancements for microbiome data analysis.