PROJECTION Algorithm for Motif Finding on GPUs

TL;DR

This paper presents a GPU-accelerated parallel implementation of the PROJECTION algorithm for motif finding in computational biology, addressing computational challenges and optimizing performance for large input sequences.

Contribution

The paper introduces a CUDA-based parallel implementation of the PROJECTION algorithm, improving efficiency and handling GPU memory constraints for large biological datasets.

Findings

Achieved significant speedup over CPU implementations.

Identified key GPU optimization strategies for motif finding.

Demonstrated effectiveness on large input sequences.

Abstract

Motif finding is one of the NP-complete problems in Computational Biology. Existing nondeterministic algorithms for motif finding do not guarantee the global optimality of results and are sensitive to initial parameters. To address this problem, the PROJECTION algorithm provides a good initial estimate that can be further refined using local optimization algorithms such as EM, MEME or Gibbs. For large enough input (600-1000 bp per sequence) or for challenging motif finding problems, the PROJECTION algorithm may run in an inordinate amount of time. In this paper we present a parallel implementation of the PROJECTION algorithm in Graphics Processing Units (GPUs) using CUDA. We also list down several major issues we have encountered including performing space optimizations because of the GPU's space limitations.