EdgeAlign

EdgeAlign performs pairwise DNA sequence alignment using a deep Q-network (DQN) deep reinforcement learning strategy optimized for execution on edge computing hardware such as microcontrollers and FPGAs.

Key Features:

• Deep Reinforcement Learning: Employs a deep Q-network (DQN) agent trained with deep reinforcement learning to make alignment decisions.
• Sliding-window alignment: Aligns fixed-length sub-sequences by scanning entire sequences with a sliding-window mechanism.
• Edge device compatibility: Optimized for stand-alone edge devices including microcontrollers and platforms demonstrated on NVIDIA Jetson Nano and Digilent Arty A7-100T with Xilinx XC7A35T Artix-7 FPGA.
• Resource efficiency: Maintains hardware resource consumption independent of sequence length.
• AutoML-based model reduction: Uses AutoML techniques to reduce neural network model size for compact deployment on constrained hardware.

Scientific Applications:

• Genome-based diagnostics: Enables DNA sequencing alignments to support identification of functional, structural, or evolutionary relationships between sequences.
• Edge-deployed diagnostics in resource-limited settings: Facilitates translation of sequencing-based diagnostic workflows to low-cost healthcare environments via on-device alignment.

Methodology:

Train a deep Q-network (DQN) agent via deep reinforcement learning to align fixed-length sub-sequences using a sliding-window across whole sequences, and apply AutoML for neural network model size reduction to enable constant resource usage on edge hardware.