GDASC

GDASC identifies hidden batch factors in high-throughput biological datasets, particularly RNA sequencing data, using a GPU-parallelized implementation of the DASC algorithm.

Key Features:

• GPU Parallelization: Leverages GPU acceleration to speed computation, achieving more than 50-fold faster runtime than the original CPU-based implementation on RNA sequencing quality-control datasets.
• DASC Algorithm: Implements the DASC algorithm for systematic detection of hidden batch factors.
• Data-Adaptive Shrinkage: Applies data-adaptive shrinkage to adjust observations and enhance batch effect detection accuracy.
• Semi-Non-Negative Matrix Factorization (SNMF): Uses SNMF to decompose datasets and reveal components corresponding to hidden batch effects.
• Parallelization Strategies: Employs parallel strategies for convex clustering solutions and matrix decomposition processes to improve speed and scalability.
• Accuracy: Maintains high accuracy in detecting batch effects critical for the integrity of downstream analyses.

Scientific Applications:

• RNA-seq Quality Control: Detects hidden batch factors to support quality control in RNA sequencing studies.
• Genomics and Transcriptomics: Identifies batch effects in genomics and transcriptomics datasets to prevent confounding of biological signals.
• Differential Gene Expression Analysis: Improves reliability of downstream analyses such as differential gene expression by identifying and enabling correction of batch effects.

Methodology:

Data-adaptive shrinkage and semi-non-negative matrix factorization (SNMF) are applied, with computation optimized via GPU parallel processing and parallel strategies for convex clustering and matrix decomposition.