IPCARF

IPCARF predicts long non-coding RNA (lncRNA)-disease associations by integrating disease semantic similarity (from two models applied to a disease directed acyclic graph), lncRNA similarity, and Gaussian nuclear similarity, then applying incremental principal component analysis (IPCA) and random forest (RF) classification to prioritize candidate associations.

Key Features:

• Disease semantic similarity: Computes a disease semantic similarity matrix using two distinct models applied to a disease directed acyclic graph (DAG).
• Similarity integration: Constructs characteristic vectors for each lncRNA-disease pair by integrating disease similarity, lncRNA similarity, and Gaussian nuclear similarity matrices.
• Dimensionality reduction: Applies incremental principal component analysis (IPCA) to reduce feature dimensionality while retaining essential information.
• Classification: Trains a random forest (RF) classifier on the reduced feature subspace to predict lncRNA-disease associations.
• Model evaluation: Uses 10-fold cross-validation to assess predictive performance, yielding an AUC of 0.8529 before parameter optimization and 0.8611 after optimization.
• Parameter optimization: Employs a grid search algorithm to select optimal model parameters.
• Comparative benchmarking: Demonstrates superior performance relative to LRLSLDA, LRLSLDA-LNCSIM, TPGLDA, NPCMF, and ncPred in comparative analyses.

Scientific Applications:

• lncRNA-disease association prediction: Prioritizes candidate lncRNA-disease links for downstream validation.
• Disease mechanism and biomarker discovery: Aids identification of lncRNA-related mechanisms and potential biomarkers for diagnosis, treatment, prognosis, and drug response prediction.
• Method benchmarking: Provides a framework for comparative evaluation against existing lncRNA-disease prediction methods.

Methodology:

Compute disease semantic similarity using two models on a disease DAG; integrate disease similarity, lncRNA similarity, and Gaussian nuclear similarity into characteristic vectors for lncRNA-disease pairs; apply incremental principal component analysis (IPCA) for dimensionality reduction; train a random forest classifier; evaluate with 10-fold cross-validation and perform grid search for parameter optimization (AUC 0.8529 before optimization, 0.8611 after).