Large-scale metagenomic assemblies have uncovered thousands of new species greatly expanding the known diversity of microbiomes in specific habitats. To investigate the roles of these uncultured species in human health or the environment, researchers need to incorporate their genome assemblies into a reference database for taxonomic classification. However, this procedure is hindered by the lack of a well-curated taxonomic tree for newly discovered species, which is required by current metagenomics tools. Here we report DeepMicrobes, a deep learning-based computational framework for taxonomic classification that allows researchers to bypass this limitation. We show the advantage of DeepMicrobes over state-of-the-art tools in species and genus identification and comparable accuracy in abundance estimation. We trained DeepMicrobes on genomes reconstructed from gut microbiomes and discovered potential novel signatures in inflammatory bowel diseases. DeepMicrobes facilitates effective investigations into the uncharacterized roles of metagenomic species.

No abstract

The current dogma in ophthalmology and vision research presumes the intraocular environment to be sterile. However, recent evidence of intestinal bacterial translocation into the bloodstream and many other internal organs including the eyes, found in healthy and diseased animal models, suggests that the intraocular cavity may also be inhabited by a microbial community. Here, we tested intraocular samples from over 1000 human eyes. Using quantitative PCR, negative staining transmission electron microscopy, direct culture, and high-throughput sequencing technologies, we demonstrated the presence of intraocular bacteria. The possibility that the microbiome from these low-biomass communities could be a contamination from other tissues and reagents was carefully evaluated and excluded. We also provide preliminary evidence that a disease-specific microbial signature characterized the intraocular environment of patients with age-related macular degeneration and glaucoma, suggesting that either spontaneous or pathogenic bacterial translocation may be associated with these common sight-threatening conditions. Furthermore, we revealed the presence of an intraocular microbiome in normal eyes from non-human mammals and demonstrated that this varied across species (rat, rabbit, pig, and macaque) and was established after birth. These findings represent the first-ever evidence of intraocular microbiota in humans.

10Taxonomic classification is a crucial step for metagenomics applications 11 including disease diagnostics, microbiome analyses, and outbreak tracing. Yet 12it is unknown what deep learning architecture can capture microbial genome-13 wide features relevant to this task. We report DeepMicrobes 14 (https://github.com/MicrobeLab/DeepMicrobes), a computational framework 15 that can perform large-scale training on > 10,000 RefSeq complete microbial 16 genomes and accurately predict the species-of-origin of whole metagenome 17 shotgun sequencing reads. We show the advantage of DeepMicrobes over 18 state-of-the-art tools in precisely identifying species from microbial community 19 sequencing data. Therefore, DeepMicrobes expands the toolbox of taxonomic 20 classification for metagenomics and enables the development of further deep 21 learning-based bioinformatics algorithms for microbial genomic sequence 22 analysis. 23 4 hypothesize that deep learning can automatically discover taxonomic 45 classification-relevant and genome-wide shared features appearing in short 46 metagenomics sequencing reads given a well-designed deep neural network 47 (DNN) architecture. 48Deep learning has made tremendous recent advances in genomics 5 . 49Taking one-hot encoded DNA sequences as input, the DNNs that have been 50 employed to genomic data fall into two major categories, convolutional neural 51 networks (CNNs) and a hybrid of CNNs and recurrent neural networks (RNNs). 52For example, DeepSEA 6 , PrimateAI 7 and SpliceAI 8 used CNNs to predict the 53 impact of genetic variation. Seq2species 9 also adopted CNNs to predict the 54 species-of-origin of 16S data. DeeperBind 10 and DanQ 11 used hybrid 55 architectures to predict transcription factor binding and DNA accessibility. 56Despite the success of these applications, it remains unknown what DNN 57 architecture and DNA encoding method are suitable for taxonomic classification 58 of metagenomics data. 59Here we describe DeepMicrobes, a k-mer embedding-based recurrent 60 network with attention mechanism (Fig. 1a). We trained the DNN on synthetic 61 reads from RefSeq complete bacterial and archaeal genomes. The first layer of 62DeepMicrobes is designed to encode k-mers to dense vectors through 63 embedding. The vectors are fed into a bidirectional long short-term memory 64 network (BiLSTM) followed by self-attention and a multilayer perceptron (MLP). 65

No abstract

Early detection of SARS-CoV-2 variants enables timely tracking of clinically important strains in order to inform the public health response. Current subtype-based variant surveillance depending on prior subtype assignment according to lag features and their continuous risk assessment may delay this process. We proposed a weighted network framework to model the frequency trajectories of mutations (FTMs) for SARS-CoV-2 variant tracing, without requiring prior subtype assignment. This framework modularizes the FTMs and conglomerates synchronous FTMs together to represent the variants. It also generates module clusters to unveil the epidemic stages and their contemporaneous variants. Eventually, the module-based variants are assessed by phylogenetic tree through sub-sampling to facilitate communication and control of the epidemic. This process was benchmarked using worldwide GISAID data, which not only demonstrated all the methodology features but also showed the module-based variant identification had highly specific and sensitive mapping with the global phylogenetic tree. When applying this process to regional data like India and South Africa for SARS-CoV-2 variant surveillance, the approach clearly elucidated the national dispersal history of the viral variants and their co-circulation pattern, and provided much earlier warning of Beta (B.1.351), Delta (B.1.617.2), and Omicron (B.1.1.529). In summary, our work showed that the weighted network modeling of FTMs enables us to rapidly and easily track down SARS-CoV-2 variants overcoming prior viral subtyping with lag features, accelerating the understanding and surveillance of COVID-19.

No abstract

No abstract