multiclassPairs

multiclassPairs implements multiclass classification of transcriptomic data using k-Top Scoring Pairs (kTSP) algorithms to derive gene-pair rules for class prediction across multiple classes.

Key Features:

• k-Top Scoring Pairs (kTSP): Uses kTSP algorithms to generate gene expression feature-pair rules for class prediction.
• Multiclass Classification: Extends binary kTSP frameworks to handle multiclass problems for tasks such as tumor subtype prediction.
• One-Versus-Rest kTSP Scheme: Implements a one-versus-rest strategy that treats each class against all others to build multiclass classifiers.
• Pair-Based Random Forest Approach: Provides a pair-based Random Forest methodology for multiclass classification using gene-pair features.
• Handling Class Imbalances: Includes strategies to manage imbalanced class distributions in training data.
• Multiplatform Training: Supports training classifiers across different measurement platforms.
• Missing Features Management: Offers options to handle missing features in test data to enable prediction with incomplete inputs.
• Visualization Tools: Produces visualizations for training and testing results to facilitate interpretation of classification outcomes.

Scientific Applications:

• Tumor subtype prediction: Enables identification and classification of multiple tumor subtypes from transcriptomic expression data using gene-pair rules.
• Multiclass transcriptomic classification: Facilitates multiclass classification tasks in transcriptomic studies, including analyses with class imbalance and missing features.

Methodology:

Implements k-Top Scoring Pairs (kTSP), including a one-versus-rest kTSP scheme, and a pair-based Random Forest approach; additionally incorporates strategies for class imbalance, training across platforms, and options for handling missing features in test data.