Real-time raw signal genomic analysis using fully integrated memristor hardware.

Area of Science:

• Genomics and Bioinformatics
• Materials Science and Engineering
• Computer Architecture and Hardware Design

Background:

• Third-generation sequencing offers portable, real-time genomic analysis but faces data processing bottlenecks.
• Current methods require basecalling and read mapping, which are computationally intensive and energy-consuming on traditional hardware.
• On-site genomic analysis is hindered by the need for efficient, low-power data processing solutions.

Purpose of the Study:

• To develop a hardware-software codesign for real-time processing of raw genomic sequencer signals.
• To overcome the limitations of von Neumann architectures for portable, on-site genomic applications.
• To demonstrate the feasibility of in-memory computing for accelerating genomic data analysis.

Main Methods:

• A memristor-based hardware-software codesign was developed to process raw sequencer signals directly in analog memory.
• Exploited intrinsic device noise for locality-sensitive hashing and implemented parallel approximate searches in content-addressable memory.
• Integrated the system onto a fully functional memristor chip for experimental validation.

Main Results:

• Achieved a 97.15% F1 score in virus raw signal mapping, demonstrating high accuracy.
• Realized a 51× speed-up and 477× energy saving compared to application-specific integrated circuits.
• Successfully showcased on-site applications including infectious disease detection and metagenomic classification.

Conclusions:

• Memristor-based in-memory computing offers a viable solution for real-time genomic data processing.
• This approach significantly enhances the efficiency and reduces the energy consumption of portable sequencing devices.
• Enables practical, on-site genomic analysis, advancing fields like infectious disease surveillance and environmental monitoring.