Artificial intelligence in metagenome-assembled genome reconstruction: Tools, pipelines, and future directions

Metagenomic sequencing has revolutionised the field of microbial ecology, as it has led to cultivation-independent exploration of complicated microbial communities. The assembly of metagenome-assembled genomes has provided genome-scale information about uncultivated microorganisms, but issues such as sequencing errors, fragmented assemblies, residual redundancy, uneven coverage, recovery of low-abundance taxa, and highly diversified taxa continue to impair the quality of these genomes. The latest achievements in artificial intelligence, particularly in machine learning and deep learning, have played a significant role in overcoming these limitations by enhancing quality control, error correction, assembly, binning, refinement, and annotation procedures. It is demonstrated that representation learning and graph-based binning methods have high strain-level resolution and can reduce contamination in complex microbial communities, whereas artificial intelligence-based assemblers and polishing tools improve base-level precision and assembly contiguity. This review synthesises traditional and artificial intelligence-based workflows involved in the reconstruction of metagenome-assembled genomes, encompassing quality control, assembly, binning, refinement, and annotation, as well as quantitative benchmarking of significant artificial intelligence-based pipelines. As future directions, the focus on emerging trends, such as explainable artificial intelligence, federated learning, cloud-native scalable pipelines, multimodal and multi-omics integration, and large language model-based annotation, is covered. In general, the incorporation of artificial intelligence represents a paradigm shift in the reconstruction of metagenome-assembled genomes, allowing for a more relevant, scalable, and biologically informative search of the microbial dark matter in various ecosystems.