Microbiomes are inherently linked by their structural similarity, yet the global features of such similarity are not clear. Here, we propose as a solution a search-based microbiome transition network. By traversing a composition-similarity-based network of 177,022 microbiomes, we show that although the compositions are distinct by habitat, each microbiome is on-average only seven neighbors from any other microbiome on Earth, indicating the inherent homology of microbiomes at the global scale. This network is scale-free, suggesting a high degree of stability and robustness in microbiome transition. By tracking the minimum spanning tree in this network, a global roadmap of microbiome dispersal was derived that tracks the potential paths of formulating and propagating microbiome diversity. Such search-based global microbiome networks, reconstructed within hours on just one computing node, provide a readily expanded reference for tracing the origin and evolution of existing or new microbiomes. IMPORTANCE It remains unclear whether and how compositional changes at the "community to community" level among microbiomes are linked to the origin and evolution of global microbiome diversity. Here we propose a microbiome transition model and a network-based analysis framework to describe and simulate the variation and dispersal of the global microbial beta-diversity across multiple habitats. The traversal of a transition network with 177,022 samples shows the inherent homology of microbiome at the global scale. Then a global roadmap of microbiome dispersal derived from the network tracks the potential paths of formulating and propagating microbiome diversity. Such search-based microbiome network provides a readily expanded reference for tracing the origin and evolution of existing or new microbiomes at the global scale.

A quantitative and objective indicator for skin health via the microbiome is of great interest for personalized skin care, but differences among skin sites and across human populations can make this goal challenging. A three-city (two Chinese and one American) comparison of skin microbiota from atopic dermatitis (AD) and healthy pediatric cohorts revealed that, although city has the greatest effect size (the skin microbiome can predict the originated city with near 100% accuracy), a microbial index of skin health (MiSH) based on 25 bacterial genera can diagnose AD with 83 to ∼95% accuracy within each city and 86.4% accuracy across cities (area under the concentration-time curve [AUC], 0.90). Moreover, nonlesional skin sites across the bodies of AD-active children (which include shank, arm, popliteal fossa, elbow, antecubital fossa, knee, neck, and axilla) harbor a distinct but lesional state-like microbiome that features relative enrichment of Staphylococcus aureus over healthy individuals, confirming the extension of microbiome dysbiosis across body surface in AD patients. Intriguingly, pretreatment MiSH classifies children with identical AD clinical symptoms into two host types with distinct microbial diversity and treatment effects of corticosteroid therapy. These findings suggest that MiSH has the potential to diagnose AD, assess risk-prone state of skin, and predict treatment response in children across human populations.IMPORTANCE MiSH, which is based on the skin microbiome, can quantitatively assess pediatric skin health across cohorts from distinct countries over large geographic distances. Moreover, the index can identify a risk-prone skin state and compare treatment effect in children, suggesting applications in diagnosis and patient stratification.

With the expansion of microbiome sequencing globally, a key challenge is to relate new microbiome samples to the existing space of microbiome samples. Here, we present Microbiome Search Engine (MSE), which enables the rapid search of query microbiome samples against a large, well-curated reference microbiome database organized by taxonomic similarity at the whole-microbiome level. Tracking the microbiome novelty score (MNS) over 8 years of microbiome depositions based on searching in more than 100,000 global 16S rRNA gene amplicon samples, we detected that the structural novelty of human microbiomes is approaching saturation and likely bounded, whereas that in environmental habitats remains 5 times higher. Via the microbiome focus index (MFI), which is derived from the MNS and microbiome attention score (MAS), we objectively track and compare the structural-novelty and attracted-attention scores of individual microbiome samples and projects, and we predict future trends in the field. For example, marine and indoor environments and mother-baby interactions are likely to receive disproportionate additional attention based on recent trends. Therefore, MNS, MAS, and MFI are proposed "alt-metrics" for evaluating a microbiome project or prospective developments in the microbiome field, both of which are done in the context of existing microbiome big data.IMPORTANCE We introduce two concepts to quantify the novelty of a microbiome. The first, the microbiome novelty score (MNS), allows identification of microbiomes that are especially different from what is already sequenced. The second, the microbiome attention score (MAS), allows identification of microbiomes that have many close neighbors, implying that considerable scientific attention is devoted to their study. By computing a microbiome focus index based on the MNS and MAS, we objectively track and compare the novelty and attention scores of individual microbiome samples and projects over time and predict future trends in the field; i.e., we work toward yielding fundamentally new microbiomes rather than filling in the details. Therefore, MNS, MAS, and MFI can serve as "alt-metrics" for evaluating a microbiome project or prospective developments in the microbiome field, both of which are done in the context of existing microbiome big data.

Microbiome-based disease classification depends on well-validated disease-specific models or a priori organismal markers. These are lacking for many diseases. Here, we present an alternative, search-based strategy for disease detection and classification, which detects diseased samples via their outlier novelty versus a database of samples from healthy subjects and then compares these to databases of samples from patients. Our strategy's precision, sensitivity, and speed outperform model-based approaches. In addition, it is more robust to platform heterogeneity and to contamination in 16S rRNA gene amplicon data sets. This search-based strategy shows promise as an important first step in microbiome big-data-based diagnosis.IMPORTANCE Here, we present a search-based strategy for disease detection and classification, which detects diseased samples via their outlier novelty versus a database of samples from healthy subjects and then compares them to databases of samples from patients. This approach enables the identification of microbiome states associated with disease even in the presence of different cohorts, multiple sequencing platforms, or significant contamination.