MSAProbs-MPI: parallel multiple sequence aligner for distributed-memory systems.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein multiple sequence alignment (MSA) is crucial for understanding protein function and evolution.
• Existing tools like MSAProbs offer high accuracy but face performance limitations with large datasets.
• Hidden Markov Models (HMMs) are foundational for accurate MSA.

Purpose of the Study:

• To develop a parallel version of MSAProbs to accelerate large-scale protein sequence alignments.
• To evaluate the performance and scalability of the new distributed-memory parallel tool.
• To compare MSAProbs-MPI against existing state-of-the-art MSA tools, including GPU-accelerated versions.

Main Methods:

• Implementation of MSAProbs using Message Passing Interface (MPI) for distributed-memory systems.
• Performance evaluation on a multicore CPU cluster (32 nodes, Intel Haswell processors).
• Benchmarking against the original MSAProbs and GPU-accelerated QuickProbs.

Main Results:

• MSAProbs-MPI achieved over a tenfold reduction in execution time for typical large-scale datasets.
• The parallel version demonstrated superior performance compared to GPU-accelerated QuickProbs on a Tesla K20.
• MSAProbs-MPI effectively processed datasets that were intractable for MSAProbs (time) and QuickProbs (memory).

Conclusions:

• MSAProbs-MPI offers a scalable and efficient solution for large-scale protein multiple sequence alignment.
• The distributed-memory parallel approach overcomes the computational bottlenecks of previous tools.
• This advancement enables more comprehensive genomic and proteomic analyses through faster and more robust MSA.