Efficient end-to-end long-read sequence mapping using minimap2-fpga integrated with hardware accelerated chaining.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Minimap2 is the standard for reference-based sequence mapping in third-generation long-read sequencing.
• Existing software, while efficient, requires further speed improvements for large-scale genomic datasets.

Purpose of the Study:

• To develop and evaluate minimap2-fpga, a hardware-accelerated version of minimap2 using Field-Programmable Gate Arrays (FPGAs).
• To address challenges in integrating FPGA acceleration, including processing time prediction, data transfer, and memory optimization for large datasets.

Main Methods:

• Integration of an FPGA kernel optimized for chaining into the minimap2 software.
• Development of strategies for predicting hardware processing times and managing overheads in a multi-threaded environment.
• Optimization of memory management for handling large, realistic sequencing datasets.

Main Results:

• Minimap2-fpga achieves significant speed-ups: up to 79% faster for Oxford Nanopore Technologies (ONT) and 53% faster for Pacific Biosciences (PacBio) datasets without base-level alignment.
• With base-level alignment, speed-ups are up to 62% for ONT and 10% for PacBio.
• Accuracy remains comparable to the original minimap2 for both data types and alignment modes.

Conclusions:

• Minimap2-FPGA provides a substantial performance enhancement for long-read sequencing mapping.
• The solution is compatible with both Intel and Xilinx FPGA systems.
• A documented library for the FPGA chaining kernel is provided for future research in sequence alignment software development.