mirPLS

mirPLS identifies microRNAs associated with cancer subtypes by modeling both linear and non-linear relationships to enable subtype discovery and biomarker selection.

Key Features:

• Partial Linear Structure Identification: mirPLS employs a Partial Linear Structure (PLS) approach to detect miRNAs with linear or non-linear associations to cancer status.
• Regularization Parameter Optimization: A cross-validation function, cv_mirPLS, determines optimal regularization parameters lambda1 (recommended range 0–1) and lambda2 (recommended range 0–10).
• miRNA Selection: mirPLS selects miRNAs based on estimated associations using the determined regularization parameters, capturing both linear and non-linear relationships.
• Spectral Clustering: A sp_clustering function performs spectral clustering on patients using selected non-linearly associated miRNAs to delineate distinct cancer subtypes.

Scientific Applications:

• Cancer Subtyping Accuracy: Identification of non-linearly associated miRNAs improves delineation of cancer subtypes compared to approaches that consider only linear associations.
• Predictive and Biological Relevance: Subtypes defined using mirPLS-selected miRNAs show increased predictive value for patient survival and greater biological interpretability.

Methodology:

mirPLS uses cross-validation (cv_mirPLS) to optimize lambda1 and lambda2, applies a Partial Linear Structure model for miRNA selection, performs spectral clustering (sp_clustering) on patients using selected non-linear miRNAs, and its performance was evaluated using simulation studies; it is applicable to miRNA datasets such as TCGA.