Interpretable machine learning identifies paediatric Systemic Lupus Erythematosus subtypes based on gene expression...

Area of Science:

• Immunology
• Computational Biology
• Genetics

Background:

• Transcriptomic analysis is standard for identifying gene expression differences in diseases like Systemic Lupus Erythematosus (SLE).
• However, traditional methods struggle to capture the complex interplay of multiple genes driving disease progression.
• Understanding these combinatorial gene effects is crucial for accurate patient stratification and treatment.

Purpose of the Study:

• To apply rule-based machine learning (RBML) models to a pediatric SLE blood expression dataset.
• To develop gene networks capable of differentiating between low (DA1) and high (DA3) disease activity states.
• To identify novel gene patterns and biological pathways involved in SLE pathogenesis.

Main Methods:

• Utilized rule-based machine learning (RBML) and rule networks (RN) on a pediatric SLE blood expression dataset.
• Applied unsupervised hierarchical clustering to identify subgroups within the data.
• Correlated identified gene sets and subgroups with clinical variables.

Main Results:

• Achieved 81% accuracy in distinguishing between low (DA1) and high (DA3) disease activity states.
• Hierarchical clustering revealed distinct subgroups associated with specific immune responses and disease flare states.
• Identified key genes involved in interferon pathways (IFI35, OTOF), SLE cell types (KLRB1), and autophagy/NF-κB pathways (CKAP4).

Conclusions:

• RBML models can effectively identify complex gene networks driving disease activity in heterogeneous conditions like SLE.
• The identified gene subgroups correlate with clinical variables, offering insights into disease progression and immune axis involvement.
• This approach facilitates patient stratification and holds potential for personalized therapeutic strategies in SLE.