RandomReadsMG: Rapid, realistic metagenome simulation at terabase scale for benchmarking and experimental design

Realistic synthetic metagenomes are essential for benchmarking bioinformatics tools, guiding experimental design, and generating labeled data for artificial intelligence and machine learning. We present RandomReadsMG, an open-source metagenomic read simulator distributed with BBTools. It generates communities containing hundreds to 50,000 genomes in a single command and supports user-defined abundance profiles, configurable within-genome coverage variation, retained read provenance, sequencing errors, and library artifacts. Presets are provided for Illumina, Oxford Nanopore Technologies (ONT), and PacBio sequencing. In benchmarks, RandomReadsMG produced terabase-scale datasets in under 6 h while maintaining bounded memory use. Simulations based on empirical drinking water profiles preserved genome-level depth after remapping. Controlled pathogen spike-ins further showed how simulation can estimate thresholds for read detection and metagenome-assembled genome (MAG) recovery. RandomReadsMG provides a fast, reproducible framework for metagenomics benchmarking, biosurveillance study design, and large-scale synthetic data generation.