DECA

DECA implements a scalable, distributed re-implementation of XHMM to detect copy-number variants (CNVs) from whole-exome sequencing data using ADAM and Apache Spark to accelerate and scale CNV discovery.

Key Features:

• Scalability: Operates across multi-core shared-memory systems and large distributed clusters using Apache Spark to handle varying dataset sizes.
• Performance optimization: Incorporates algorithmic optimizations that eliminate unnecessary computations, yielding measured speedups such as CNV discovery from a read-depth matrix in 9.3 minutes on a 16-core workstation (35.3× faster than XHMM) and 12.7 minutes using 10 executor cores on a Spark cluster (18.8× speedup).
• Large-scale data handling: Processes CNV discovery from original BAM files at cluster scale, demonstrated as 292 minutes using 640 executor cores on a Spark cluster.
• Algorithmic enhancements: Integrates improvements to optimize computational resource use for genome-wide analyses and high-throughput CNV discovery.
• Research-focused configuration exploration: Enables exploration of a broader configuration parameter space to fine-tune CNV detection across large cohorts.

Scientific Applications:

• Whole-exome CNV discovery: Detection of copy-number variants from whole-exome sequencing read-depth matrices and BAM files.
• Large-cohort genomic studies: High-throughput CNV analysis across large sample cohorts enabled by distributed computation.
• Genome-wide analyses and parameter optimization: Genome-wide CNV analyses with the ability to explore configuration parameter space for method tuning.

Methodology:

Re-implements the XHMM algorithm using ADAM and Apache Spark with parallelization and algorithmic optimizations to distribute computation across multi-core workstations and Spark clusters, operating on read-depth matrices and original BAM files.