DeepSimulator

DeepSimulator simulates Nanopore sequencing electrical current signals and produces corresponding reads using deep learning to reproduce realistic Nanopore sequencing data for algorithm development and benchmarking.

Key Features:

• Sequence generator: Updates the sample read length distribution to produce simulated sequences that reflect current Nanopore read length characteristics.
• Signal generation: Uses a revamped signal generation module with a context-independent pore model and applies a low-pass filter in post-processing to produce electrical current signals that mimic real Nanopore data.
• Basecalling: Integrates support for the Guppy basecaller operating on GPU and CPU to convert simulated signals into nucleotide reads.
• Computational optimizations: Implements improved multiprocessing control and refined memory and storage management to optimize simulation performance.

Scientific Applications:

• Algorithm development: Provides realistic simulated reads and signals for development and testing of sequence analysis algorithms.
• Machine learning training: Supplies synthetic datasets that closely resemble Nanopore outputs for training and validating machine learning models.
• Comparative analysis: Enables comparative studies between simulated and actual Nanopore data to investigate sequencing artifacts and biases.

Methodology:

Takes a reference genome or assembled contigs as input, generates sequences with an updated read length distribution, simulates electrical current signals using deep learning-based pore models (context-independent), applies a low-pass filter for signal post-processing, and performs basecalling with Guppy on GPU/CPU while employing multiprocessing, memory, and storage optimizations.