OPLS-Based Multiclass Classification and Data-Driven Interclass Relationship Discovery.

Area of Science:

• Omics sciences
• Drug discovery
• Clinical research

Background:

• Multiclass datasets are prevalent in modern scientific research, posing challenges for analysis.
• Existing two-class OPLS-DA models are effective but difficult to apply to multiclass problems, often requiring manual, time-consuming transformations.

Purpose of the Study:

• To introduce Orthogonal Partial Least Squares-Hierarchical Discriminant Analysis (OPLS-HDA) for data-driven multiclass classification.
• To provide an efficient and interpretable method for dissecting complex multiclass data.

Main Methods:

• OPLS-HDA integrates Hierarchical Cluster Analysis (HCA) with the OPLS-DA framework.
• A decision tree approach is employed for multiclass classification and visualization of interclass relationships.
• Cross-validation is utilized to prevent overfitting and ensure predictive reliability.

Main Results:

• OPLS-HDA demonstrates competitive performance across diverse datasets compared to eight established methods.
• The method provides intuitive visualization of interclass relationships.
• Benchmark results confirm OPLS-HDA's effectiveness in multiclass data analysis.

Conclusions:

• OPLS-HDA is a significant advancement for multiclass data analysis, offering versatility, interpretability, and ease of use.
• This method provides a powerful tool for researchers in omics, drug discovery, and clinical research.
• OPLS-HDA addresses key challenges in handling and interpreting complex, large-scale multiclass datasets.