PanClassif

PanClassif classifies cancer from bulk and single-cell RNA-seq gene expression data to detect malignancy and predict cancer types.

Key Features:

• Efficient Gene Selection: Selects a minimal and informative set of genes using an ANOVA-based test for downstream classification.
• Noise Handling with k-NN Smoothing: Preprocesses RNA-seq data with k-Nearest Neighbor (k-NN) smoothing to mitigate technical and biological noise.
• Data Balancing with SMOTE: Balances training data by generating synthetic samples for underrepresented classes using the Synthetic Minority Over-sampling Technique (SMOTE).
• Comprehensive Evaluation Across Classifiers: Evaluated diverse machine learning classifiers on 22 TCGA cancer types comprising 8,287 cancer and 680 normal samples, covering binary cancer prediction and multi-class cancer classification tasks.
• Versatility Across Datasets: Validated on two single-cell RNA-seq datasets from the Gene Expression Omnibus (GEO).

Scientific Applications:

• Cancer type classification: Classifies cancer types based on bulk and single-cell RNA-seq gene expression data.
• Research and clinical support: Enhances classifier performance for academic research and clinical applications where accurate classification can inform treatment strategies and improve patient outcomes.

Methodology:

Applies k-Nearest Neighbor (k-NN) smoothing to preprocess RNA-seq data, selects genes using an ANOVA-based test, balances data with SMOTE, and trains diverse machine learning classifiers.