Bronwyn L. MacInnis

Pathogen genomic analysis is central to tracking, understanding and containing outbreaks ^1–13 , but the complexity and cost of state-of-the-art phylogenetic tools limit global access and impact. Here we introduce Delphy, an exact reformulation of Bayesian phylogenetics ^14–17 designed to transform its speed, scalability and accessibility while retaining Bayesian state-of-the-art accuracy. Delphy’s central data structure, an explicit mutation-annotated tree, takes advantage of the high sequence similarity of large-scale epidemic datasets ^18–20 for efficient tree exploration and convergence. By reproducing key analyses from recent major epidemics, including Ebola ^1,21 , Zika ² , SARS-CoV-2 (ref. ²² ), mpox ^3,4 and H5N1 (refs. ^23,24 ), we demonstrate state-of-the-art accuracy with up to 2–3 orders of magnitude improvements in speed. Assessing Delphy’s scalability, we show that a simulated dataset of 100,000 sequences can be analysed within a day. We distribute Delphy as a client-side web application that enables local, interactive analysis of raw data on the user’s machine. Delphy automatically identifies key viral lineages and mutations, as well as their emergence and prevalence through time, with quantified uncertainties grounded in Bayesian theory. Delphy establishes Bayesian phylogenetics as a fast, accessible frontline tool for future outbreak response.

Understanding respiratory virus transmission requires integrating pathogen genomic data with detailed epidemiological context. We analyzed >85,000 SARS-CoV-2 genomes linked to epidemiological data collected in Massachusetts from November 2021 to January 2023. Collected in 666 settings, including schools, colleges, and skilled nursing facilities, this dataset enabled direct comparison of within-setting transmission to community-wide risk. Elevated transmission was concentrated in specific age-defined subpopulations, older adolescents in schools, undergraduates in colleges, and residents in nursing facilities, with little additional risk among other groups, notably including staff. Young adults, particularly in colleges, played a disproportionate role in spread, and new viral lineages consistently expanded first within this group. Viral spread followed reproducible urban-to-rural gradients, and vaccination was associated with reduced onward transmission. We further estimated sequencing thresholds required for timely detection of emerging variants. Together, these findings reveal consistent demographic and spatial structure in SARS-CoV-2 transmission that can inform future viral respiratory pathogen genomic surveillance.

Transmission reconstruction––the inference of who infects whom in disease outbreaks––offers critical insights into how pathogens spread and provides opportunities for targeted control measures. We developed JUNIPER (Joint Underlying Network Inference for Phylogenetic and Epidemiological Reconstructions), a highly-scalable pathogen outbreak reconstruction tool that incorporates intrahost variation, incomplete sampling, and algorithmic parallelization. Central to JUNIPER is a statistical model for within-host variant frequencies observed by next generation sequencing, which we validated on a dataset of over 160,000 deep-sequenced SARS-CoV-2 genomes. Combining this within-host variation model with population-level evolutionary and transmission models, we developed a method for inferring phylogenies and transmission trees simultaneously. We benchmarked JUNIPER on computer-generated and real outbreaks in which transmission links were known or epidemiologically confirmed. We demonstrated JUNIPER’s real-world utility on two large-scale datasets: over 1,500 bovine H5N1 cases and over 13,000 human COVID-19 cases. Based on these analyses, we quantified the elevated H5N1 transmission rates in California and identified high-confidence transmission events, and demonstrated the efficacy of vaccination for reducing SARS-CoV-2 transmission. By overcoming computational and methodological limitations in existing outbreak reconstruction tools, JUNIPER provides a robust framework for studying pathogen spread at scale.

Pathogen genomic analysis is central to tracking, understanding, and containing outbreaks, but complexity and high costs of state-of-the-art (SOTA) phylogenetic tools limit global access and impact. We introduce Delphy, an exact reformulation of Bayesian phylogenetics designed to transform its speed, scalability and accessibility while retaining SOTA accuracy. Delphy's central data structure, an Explicit Mutation Annotated Tree, exploits the high sequence similarity in large-scale epidemic datasets for efficient tree exploration and convergence. By reproducing key analyses from recent major epidemics (Ebola, Zika, SARS-CoV-2, mpox, and H5N1), we demonstrate SOTA accuracy with up to 1,000x speedups. Assessing Delphy's scalability, we show that a simulated dataset of 100,000 sequences can be analyzed in under a day—the largest such computation to date. We distribute Delphy as a client-side web application, enabling users worldwide to turn raw data into interactive results within minutes, without the data ever leaving the user's machine. Delphy automatically identifies key viral lineages and mutations, as well as their emergence and prevalence through time, all with quantified uncertainties derived from a solid theoretical foundation. Delphy shows the power of Bayesian phylogenetics as a fast, accessible frontline tool for tackling future outbreaks.

Transmission reconstruction--the inference of who infects whom in disease outbreaks--offers critical insights into how pathogens spread and provides opportunities for targeted control measures. We developed JUNIPER (Joint Underlying Network Inference for Phylogenetic and Epidemiological Reconstructions), a highly-scalable pathogen outbreak reconstruction tool that incorporates intrahost variation, incomplete sampling, and algorithmic parallelization. Central to JUNIPER is a statistical model for within-host variant frequencies observed by next generation sequencing, which we validated on a dataset of over 160,000 deep-sequenced SARS-CoV-2 genomes. Combining this within-host variation model with population-level evolutionary and transmission models, we developed a method for inferring phylogenies and transmission trees simultaneously. We benchmarked JUNIPER on computer-generated and real outbreaks in which transmission links were known or epidemiologically confirmed. We demonstrated JUNIPER's real-world utility on two large-scale datasets: over 1,500 bovine H5N1 cases and over 13,000 human COVID-19 cases. Based on these analyses, we quantified the elevated H5N1 transmission rates in California and identified high-confidence transmission events, and demonstrated the efficacy of vaccination for reducing SARS-CoV-2 transmission. By overcoming computational and methodological limitations in existing outbreak reconstruction tools, JUNIPER provides a robust framework for studying pathogen spread at scale.

Highly Pathogenic Avian Influenza strain H5N1 has caused a multi-state outbreak among US dairy cattle, spreading across 15 states and infecting hundreds of herds since its onset. We rapidly developed and optimized PCR-based detection assays and sequencing protocols to support H5N1 molecular surveillance. Using 214 retail milk from 20 states for methods development, we found that H5N1 concentrations by digital PCR strongly correlated with qPCR cycle threshold (Ct) values, with dPCR exhibiting greater sensitivity. We also found that metagenomic sequencing after hybrid selection was best for higher concentration samples while amplicon sequencing performs best for lower concentrations. By establishing these methods, we were able to support the creation of a statewide surveillance program to test bulk milk samples monthly from all cattle dairy farms within Massachusetts, which remain negative to date. The methods, workflow, and recommendations described here provide a framework for others aiming to conduct H5N1 surveillance efforts.

The worldwide decline in malaria incidence is revealing the extensive burden of non-malarial febrile illness (NMFI), which remains poorly understood and difficult to diagnose. To characterize NMFI in Senegal, we collected venous blood and clinical metadata in a cross-sectional study of febrile patients and healthy controls in a low malaria burden area. Using 16S and untargeted sequencing, we detected viral, bacterial, or eukaryotic pathogens in 23% (38/163) of NMFI cases. Bacteria were the most common, with relapsing feverBorreliaand spotted feverRickettsiafound in 15.5% and 3.8% of cases, respectively. Four viral pathogens were found in a total of 7 febrile cases (3.5%). Sequencing also detected undiagnosedPlasmodium, including one putativeP. ovaleinfection. We developed a logistic regression model that can distinguishBorreliafrom NMFIs with similar presentation based on symptoms and vital signs (F1 score: 0.823). These results highlight the challenge and importance of improved diagnostics, especially forBorrelia, to support diagnosis and surveillance.

We here analyze data from the first year of an ongoing nationwide program of genetic surveillance of Plasmodium falciparum parasites in Senegal. The analysis is based on 1097 samples collected at health facilities during passive malaria case detection in 2019; it provides a baseline for analyzing parasite genetic metrics as they vary over time and geographic space. The study’s goal was to identify genetic metrics that were informative about transmission intensity and other aspects of transmission dynamics, focusing on measures of genetic relatedness between parasites. We found the best genetic proxy for local malaria incidence to be the proportion of polygenomic infections (those with multiple genetically distinct parasites), although this relationship broke down at low incidence. The proportion of related parasites was less correlated with incidence while local genetic diversity was uninformative. The type of relatedness could discriminate local transmission patterns: two nearby areas had similarly high fractions of relatives, but one was dominated by clones and the other by outcrossed relatives. Throughout Senegal, 58% of related parasites belonged to a single network of relatives, within which parasites were enriched for shared haplotypes at known and suspected drug resistance loci and at one novel locus, reflective of ongoing selection pressure.

Genome sequencing can offer critical insight into pathogen spread in viral outbreaks, but existing transmission inference methods use simplistic evolutionary models and only incorporate a portion of available genetic data. Here, we develop a robust evolutionary model for transmission reconstruction that tracks the genetic composition of within-host viral populations over time and the lineages transmitted between hosts. We confirm that our model reliably describes within-host variant frequencies in a dataset of 134,682 SARS-CoV-2 deep-sequenced genomes from Massachusetts, USA. We then demonstrate that our reconstruction approach infers transmissions more accurately than two leading methods on synthetic data, as well as in a controlled outbreak of bovine respiratory syncytial virus and an epidemiologically-investigated SARS-CoV-2 outbreak in South Africa. Finally, we apply our transmission reconstruction tool to 5,692 outbreaks among the 134,682 Massachusetts genomes. Our methods and results demonstrate the utility of within-host variation for transmission inference of SARS-CoV-2 and other pathogens, and provide an adaptable mathematical framework for tracking within-host evolution.

Parasite genetic surveillance has the potential to play an important role in malaria control. We describe here an analysis of data from the first year of an ongoing, nationwide program of genetic surveillance of Plasmodium falciparum parasites in Senegal, intended to provide actionable information for malaria control efforts. Looking for a good proxy for local malaria incidence, we found that the best predictor was the proportion of polygenomic infections (those with multiple genetically distinct parasites), although that relationship broke down in very low incidence settings (r = 0.77 overall). The proportion of closely related parasites in a site was more weakly correlated ( r = -0.44) with incidence while the local genetic diversity was uninformative. Study of related parasites indicated their potential for discriminating local transmission patterns: two nearby study areas had similarly high fractions of relatives, but one area was dominated by clones and the other by outcrossed relatives. Throughout the country, 58% of related parasites proved to belong to a single network of relatives, within which parasites were enriched for shared haplotypes at known and suspected drug resistance loci as well as at one novel locus, reflective of ongoing selection pressure.