RUBICON: a framework for designing efficient deep learning-based genomic basecallers.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Nanopore sequencing produces noisy electrical signals requiring basecalling for DNA analysis.
• Basecalling accuracy is crucial for downstream genomic interpretation.
• Current basecalling methods face challenges in computational and memory efficiency.

Purpose of the Study:

• To introduce RUBICON, a novel framework for developing efficient, hardware-optimized basecallers.
• To address the need for reduced computational and memory costs in basecalling without compromising accuracy.
• To demonstrate a new approach for accelerating nanopore sequencing data analysis.

Main Methods:

• Development of the RUBICON framework for hardware-accelerated basecaller design.
• Implementation of RUBICALL, the first mixed-precision, hardware-optimized basecaller.
• Performance evaluation against state-of-the-art basecalling algorithms.

Main Results:

• RUBICALL demonstrates significant improvements in computational and memory efficiency.
• The developed basecaller achieves superior performance compared to existing methods.
• RUBICON facilitates the creation of optimized basecalling tools for nanopore sequencing.

Conclusions:

• RUBICON provides a viable framework for creating efficient, hardware-optimized basecallers.
• RUBICALL represents a significant advancement in basecalling technology for nanopore sequencing.
• This work paves the way for future innovations in accelerating genomic data analysis.