eccCL: parallelized GPU implementation of Ensemble Classifier Chains.

Area of Science:

• Computational biology
• Machine learning
• Bioinformatics

Background:

• Multi-label classification is crucial for biomedical applications like protein function prediction and HIV drug resistance testing.
• Classifier Chains, particularly Ensemble Classifier Chains, improve prediction accuracy but struggle with computational efficiency on large datasets.
• Next-generation sequencing experiments generate massive data, demanding computationally efficient classification methods.

Purpose of the Study:

• To develop a computationally efficient implementation of Classifier Chains and Ensemble Classifier Chains.
• To leverage graphics processing units (GPUs) for parallelized computations to address the scalability challenge.
• To provide an accessible R-package for easy GPU utilization.

Main Methods:

• Developed eccCL, a parallelized and optimized GPU implementation using OpenCL.
• Integrated eccCL into an R-package for user-friendly parallelized GPU computation.
• Focused on adapting existing algorithms for efficient GPU parallelization.

Main Results:

• Achieved significant computational speed-up through GPU parallelization.
• The eccCL implementation processes over 25,000 instances per second.
• Provided an R-package enabling straightforward GPU acceleration for Classifier Chains.

Conclusions:

• eccCL offers an efficient GPU implementation for Classifier Chains, suitable for high-throughput experiments.
• The developed software addresses the computational bottleneck in large-scale multi-label classification.
• The R-package enhances accessibility for researchers utilizing GPU-accelerated machine learning.