From components to communities: bringing network science to clustering for molecular epidemiology.

Defining clusters of epidemiologically related infections is a common problem in the surveillance of infectious disease. A popular method for generating clusters is pairwise distance clustering, which assigns pairs of sequences to the same cluster if their genetic distance falls below some threshold. The result is often represented as a network or graph of nodes. A connected component is a set of interconnected nodes in a graph that are not connected to any other node. The prevailing approach to pairwise clustering is to map clusters to the connected components of the graph on a one-to-one basis. We propose that this definition of clusters is unnecessarily rigid. For instance, the connected components can collapse into one cluster by the addition of a single sequence that bridges nodes in the respective components. Moreover, the distance thresholds typically used for viruses like HIV-1 tend to exclude a large proportion of new sequences, making it difficult to train models for predicting cluster growth. These issues may be resolved by revisiting how we define clusters from genetic distances. Community detection is a promising class of clustering methods from the field of network science. A community is a set of nodes that are more densely inter-connected relative to the number of their connections to external nodes. Thus, a connected component may be partitioned into two or more communities. Here we describe community detection methods in the context of genetic clustering for epidemiology, demonstrate how a popular method (Markov clustering) enables us to resolve variation in transmission rates within a giant connected component of HIV-1 sequences, and identify current challenges and directions for further work.

SUP: a probabilistic framework to propagate genome sequence uncertainty, with applications.

Genetic sequencing is subject to many different types of errors, but most analyses treat the resultant sequences as if they are known without error. Next generation sequencing methods rely on significantly larger numbers of reads than previous sequencing methods in exchange for a loss of accuracy in each individual read. Still, the coverage of such machines is imperfect and leaves uncertainty in many of the base calls. In this work, we demonstrate that the uncertainty in sequencing techniques will affect downstream analysis and propose a straightforward method to propagate the uncertainty. Our method (which we have dubbed Sequence Uncertainty Propagation, or SUP) uses a probabilistic matrix representation of individual sequences which incorporates base quality scores as a measure of uncertainty that naturally lead to resampling and replication as a framework for uncertainty propagation. With the matrix representation, resampling possible base calls according to quality scores provides a bootstrap- or prior distribution-like first step towards genetic analysis. Analyses based on these re-sampled sequences will include a more complete evaluation of the error involved in such analyses. We demonstrate our resampling method on SARS-CoV-2 data. The resampling procedures add a linear computational cost to the analyses, but the large impact on the variance in downstream estimates makes it clear that ignoring this uncertainty may lead to overly confident conclusions. We show that SARS-CoV-2 lineage designations via Pangolin are much less certain than the bootstrap support reported by Pangolin would imply and the clock rate estimates for SARS-CoV-2 are much more variable than reported.

Optimized phylogenetic clustering of HIV-1 sequence data for public health applications.

Clusters of genetically similar infections suggest rapid transmission and may indicate priorities for public health action or reveal underlying epidemiological processes. However, clusters often require user-defined thresholds and are sensitive to non-epidemiological factors, such as non-random sampling. Consequently the ideal threshold for public health applications varies substantially across settings. Here, we show a method which selects optimal thresholds for phylogenetic (subset tree) clustering based on population. We evaluated this method on HIV-1 pol datasets (n = 14, 221 sequences) from four sites in USA (Tennessee, Washington), Canada (Northern Alberta) and China (Beijing). Clusters were defined by tips descending from an ancestral node (with a minimum bootstrap support of 95%) through a series of branches, each with a length below a given threshold. Next, we used pplacer to graft new cases to the fixed tree by maximum likelihood. We evaluated the effect of varying branch-length thresholds on cluster growth as a count outcome by fitting two Poisson regression models: a null model that predicts growth from cluster size, and an alternative model that includes mean collection date as an additional covariate. The alternative model was favoured by AIC across most thresholds, with optimal (greatest difference in AIC) thresholds ranging 0.007-0.013 across sites. The range of optimal thresholds was more variable when re-sampling 80% of the data by location (IQR 0.008 - 0.016, n = 100 replicates). Our results use prospective phylogenetic cluster growth and suggest that there is more variation in effective thresholds for public health than those typically used in clustering studies.

CoVizu: Rapid analysis and visualization of the global diversity of SARS-CoV-2 genomes.

Phylogenetics has played a pivotal role in the genomic epidemiology of severe acute respiratory syndrome coronavirus 2, such as tracking the emergence and global spread of variants and scientific communication. However, the rapid accumulation of genomic data from around the world-with over two million genomes currently available in the Global Initiative on Sharing All Influenza Data database-is testing the limits of standard phylogenetic methods. Here, we describe a new approach to rapidly analyze and visualize large numbers of SARS-CoV-2 genomes. Using Python, genomes are filtered for problematic sites, incomplete coverage, and excessive divergence from a strict molecular clock. All differences from the reference genome, including indels, are extracted using minimap2 and compactly stored as a set of features for each genome. For each Pango lineage (https://cov-lineages.org), we collapse genomes with identical features into 'variants', generate 100 bootstrap samples of the feature set union to generate weights, and compute the symmetric differences between the weighted feature sets for every pair of variants. The resulting distance matrices are used to generate neighbor-joining trees in RapidNJ that are converted into a majority-rule consensus tree for each lineage. Branches with support values below 50 per cent or mean lengths below 0.5 differences are collapsed, and tip labels on affected branches are mapped to internal nodes as directly sampled ancestral variants. Currently, we process about 2 million genomes in approximately 9 h on 52 cores. The resulting trees are visualized using the JavaScript framework D3.js as 'beadplots', in which variants are represented by horizontal line segments, annotated with beads representing samples by collection date. Variants are linked by vertical edges to represent branches in the consensus tree. These visualizations are published at https://filogeneti.ca/CoVizu. All source code was released under an MIT license at https://github.com/PoonLab/covizu.

Public health in genetic spaces: a statistical framework to optimize cluster-based outbreak detection.

Genetic clustering is a popular method for characterizing variation in transmission rates for rapidly evolving viruses, and could potentially be used to detect outbreaks in 'near real time'. However, the statistical properties of clustering are poorly understood in this context, and there are no objective guidelines for setting clustering criteria. Here, we develop a new statistical framework to optimize a genetic clustering method based on the ability to forecast new cases. We analysed the pairwise Tamura-Nei (TN93) genetic distances for anonymized HIV-1 subtype B pol sequences from Seattle (n = 1,653) and Middle Tennessee, USA (n = 2,779), and northern Alberta, Canada (n = 809). Under varying TN93 thresholds, we fit two models to the distributions of new cases relative to clusters of known cases: 1, a null model that assumes cluster growth is strictly proportional to cluster size, i.e. no variation in transmission rates among individuals; and 2, a weighted model that incorporates individual-level covariates, such as recency of diagnosis. The optimal threshold maximizes the difference in information loss between models, where covariates are used most effectively. Optimal TN93 thresholds varied substantially between data sets, e.g. 0.0104 in Alberta and 0.016 in Seattle and Tennessee, such that the optimum for one population would potentially misdirect prevention efforts in another. For a given population, the range of thresholds where the weighted model conferred greater predictive accuracy tended to be narrow (±0.005 units), and the optimal threshold tended to be stable over time. Our framework also indicated that variation in the recency of HIV diagnosis among clusters was significantly more predictive of new cases than sample collection dates (ΔAIC > 50). These results suggest that one cannot rely on historical precedence or convention to configure genetic clustering methods for public health applications, especially when translating methods between settings of low-level and generalized epidemics. Our framework not only enables investigators to calibrate a clustering method to a specific public health setting, but also provides a variable selection procedure to evaluate different predictive models of cluster growth.

Comparative analysis of HIV sequences in real time for public health.

PURPOSE OF REVIEW: The purpose of this study is to summarize recent advances in public health applications of comparative methods for HIV-1 sequence analysis in real time, including genetic clustering methods. RECENT FINDINGS: Over the past 2 years, several groups have reported the deployment of established genetic clustering methods to guide public health decisions for HIV prevention in 'near real time'. However, it remains unresolved how well the readouts of comparative methods like clusters translate to events that are actionable for public health. A small number of recent studies have begun to elucidate the linkage between clusters and HIV-1 incidence, whereas others continue to refine and develop new comparative methods for such applications. SUMMARY: Although the use of established methods to cluster HIV-1 sequence databases has become a widespread activity, there remains a critical gap between clusters and public health value.