Detecting critical slowing down in high-dimensional epidemiological systems.

Despite medical advances, the emergence and re-emergence of infectious diseases continue to pose a public health threat. Low-dimensional epidemiological models predict that epidemic transitions are preceded by the phenomenon of critical slowing down (CSD). This has raised the possibility of anticipating disease (re-)emergence using CSD-based early-warning signals (EWS), which are statistical moments estimated from time series data. For EWS to be useful at detecting future (re-)emergence, CSD needs to be a generic (model-independent) feature of epidemiological dynamics irrespective of system complexity. Currently, it is unclear whether the predictions of CSD-derived from simple, low-dimensional systems-pertain to real systems, which are high-dimensional. To assess the generality of CSD, we carried out a simulation study of a hierarchy of models, with increasing structural complexity and dimensionality, for a measles-like infectious disease. Our five models included: i) a nonseasonal homogeneous Susceptible-Exposed-Infectious-Recovered (SEIR) model, ii) a homogeneous SEIR model with seasonality in transmission, iii) an age-structured SEIR model, iv) a multiplex network-based model (Mplex) and v) an agent-based simulator (FRED). All models were parameterised to have a herd-immunity immunization threshold of around 90% coverage, and underwent a linear decrease in vaccine uptake, from 92% to 70% over 15 years. We found evidence of CSD prior to disease re-emergence in all models. We also evaluated the performance of seven EWS: the autocorrelation, coefficient of variation, index of dispersion, kurtosis, mean, skewness, variance. Performance was scored using the Area Under the ROC Curve (AUC) statistic. The best performing EWS were the mean and variance, with AUC > 0.75 one year before the estimated transition time. These two, along with the autocorrelation and index of dispersion, are promising candidate EWS for detecting disease emergence.

Forecasted Size of Measles Outbreaks Associated With Vaccination Exemptions for Schoolchildren.

Importance: Vaccine exemptions, which allow unvaccinated children to attend school, have increased by a factor of 28 since 2003 in Texas. Geographic clustering of unvaccinated children facilitates the spread of measles introductions, but the potential size of outbreaks is unclear. Objective: To forecast the range of measles outbreak sizes in each metropolitan area of Texas at 2018 and future reduced school vaccination rates. Design, Setting, and Participants: An agent-based decision analytical model using a synthetic population of Texas, derived from the 2010 US Census, was used to simulate measles transmission in the Texas population. Real schools were represented in the simulations, and the 2018 vaccination rate of each real school was applied to a simulated hypothetical equivalent. Single cases of measles were introduced, daily activities and interactions were modeled for each population member, and the number of infections over the course of 9 months was counted for 1000 simulated runs in each Texas metropolitan area. Interventions: To determine the outcomes of further decreases in vaccination coverage, additional simulations were performed with vaccination rates reduced by 1% to 10% in schools with populations that are currently undervaccinated. Main Outcomes and Measures: Expected distributions of outbreak sizes in each metropolitan area of Texas at 2018 and reduced vaccination rates. Results: At 2018 vaccination rates, the median number of cases in large metropolitan areas was typically small, ranging from 1 to 3 cases, which is consistent with outbreaks in Texas 2006 to 2017. However, the upper limit of the distribution of plausible outbreaks (the 95th percentile, associated with 1 in 20 measles introductions) exceeded 400 cases in both the Austin and Dallas metropolitan areas, similar to the largest US outbreaks since measles was eliminated in 2000. Decreases in vaccination rates in schools with undervaccinated populations in 2018 were associated with exponential increases in the potential size of outbreaks: a 5% decrease in vaccination rate was associated with a 40% to 4000% increase in potential outbreak size, depending on the metropolitan area. A mean (SD) of 64% (11%) of cases occurred in students for whom a vaccine had been refused, but a mean (SD) of 36% (11%) occurred in others (ie, bystanders). Conclusions and Relevance: This study suggests that vaccination rates in some Texas schools are currently low enough to allow large measles outbreaks. Further decreases are associated with dramatic increases in the probability of large outbreaks. Limiting vaccine exemptions could be associated with a decrease in the risk of large measles outbreaks.

Prevention of Hepatitis C by Screening and Treatment in U.S. Prisons.

BACKGROUND: The prevalence of hepatitis C virus (HCV) in U.S. prisoners is high; however, HCV testing and treatment are rare. Infected inmates released back into society contribute to the spread of HCV in the general population. Routine hepatitis screening of inmates followed by new therapies may reduce ongoing HCV transmission. OBJECTIVE: To evaluate the health and economic effect of HCV screening and treatment in prisons on the HCV epidemic in society. DESIGN: Agent-based microsimulation model of HCV transmission and progression of HCV disease. DATA SOURCES: Published literature. TARGET POPULATION: Population in U.S. prisons and general community. TIME HORIZON: 30 years. PERSPECTIVE: Societal. INTERVENTIONS: Risk-based and universal opt-out hepatitis C screening in prisons, followed by treatment in a portion of patients. OUTCOME MEASURES: Prevention of HCV transmission and associated disease in prisons and society, costs, quality-adjusted life-years (QALYs), incremental cost-effectiveness ratio (ICER), and total prison budget. RESULTS OF BASE-CASE ANALYSIS: Implementing risk-based and opt-out screening could diagnose 41,900 to 122,700 new HCV cases in prisons in the next 30 years. Compared with no screening, these scenarios could prevent 5500 to 12,700 new HCV infections caused by released inmates, wherein about 90% of averted infections would have occurred outside of prisons. Screening could also prevent 4200 to 11,700 liver-related deaths. The ICERs of screening scenarios were $19,600 to $29,200 per QALY, and the respective first-year prison budget was $900 to $1150 million. Prisons would require an additional 12.4% of their current health care budget to implement such interventions. RESULTS OF SENSITIVITY ANALYSIS: Results were sensitive to the time horizon, and ICERs otherwise remained less than $50,000 per QALY. LIMITATION: Data on transmission network, reinfection rate, and opt-out HCV screening rate are lacking. CONCLUSION: Universal opt-out HCV screening in prisons is highly cost-effective and would reduce HCV transmission and HCV-associated diseases primarily in the outside community. Investing in U.S. prisons to manage hepatitis C is a strategic approach to address the current epidemic. PRIMARY FUNDING SOURCE: National Institutes of Health.

Is population structure sufficient to generate area-level inequalities in influenza rates? An examination using agent-based models.

BACKGROUND: In New Haven County, CT (NHC), influenza hospitalization rates have been shown to increase with census tract poverty in multiple influenza seasons. Though multiple factors have been hypothesized to cause these inequalities, including population structure, differential vaccine uptake, and differential access to healthcare, the impact of each in generating observed inequalities remains unknown. We can design interventions targeting factors with the greatest explanatory power if we quantify the proportion of observed inequalities that hypothesized factors are able to generate. Here, we ask if population structure is sufficient to generate the observed area-level inequalities in NHC. To our knowledge, this is the first use of simulation models to examine the causes of differential poverty-related influenza rates. METHODS: Using agent-based models with a census-informed, realistic representation of household size, age-structure, population density in NHC census tracts, and contact rates in workplaces, schools, households, and neighborhoods, we measured poverty-related differential influenza attack rates over the course of an epidemic with a 23 % overall clinical attack rate. We examined the role of asthma prevalence rates as well as individual contact rates and infection susceptibility in generating observed area-level influenza inequalities. RESULTS: Simulated attack rates (AR) among adults increased with census tract poverty level (F = 30.5; P < 0.001) in an epidemic caused by a virus similar to A (H1N1) pdm09. We detected a steeper, earlier influenza rate increase in high-poverty census tracts-a finding that we corroborate with a temporal analysis of NHC surveillance data during the 2009 H1N1 pandemic. The ratio of the simulated adult AR in the highest- to lowest-poverty tracts was 33 % of the ratio observed in surveillance data. Increasing individual contact rates in the neighborhood did not increase simulated area-level inequalities. When we modified individual susceptibility such that it was inversely proportional to household income, inequalities in AR between high- and low-poverty census tracts were comparable to those observed in reality. DISCUSSION: To our knowledge, this is the first study to use simulations to probe the causes of observed inequalities in influenza disease patterns. Knowledge of the causes and their relative explanatory power will allow us to design interventions that have the greatest impact on reducing inequalities. CONCLUSION: Differential exposure due to population structure in our realistic simulation model explains a third of the observed inequality. Differential susceptibility to disease due to prevailing chronic conditions, vaccine uptake, and smoking should be considered in future models in order to quantify the role of additional factors in generating influenza inequalities.

FRED (a Framework for Reconstructing Epidemic Dynamics): an open-source software system for modeling infectious diseases and control strategies using census-based populations.

BACKGROUND: Mathematical and computational models provide valuable tools that help public health planners to evaluate competing health interventions, especially for novel circumstances that cannot be examined through observational or controlled studies, such as pandemic influenza. The spread of diseases like influenza depends on the mixing patterns within the population, and these mixing patterns depend in part on local factors including the spatial distribution and age structure of the population, the distribution of size and composition of households, employment status and commuting patterns of adults, and the size and age structure of schools. Finally, public health planners must take into account the health behavior patterns of the population, patterns that often vary according to socioeconomic factors such as race, household income, and education levels. RESULTS: FRED (a Framework for Reconstructing Epidemic Dynamics) is a freely available open-source agent-based modeling system based closely on models used in previously published studies of pandemic influenza. This version of FRED uses open-access census-based synthetic populations that capture the demographic and geographic heterogeneities of the population, including realistic household, school, and workplace social networks. FRED epidemic models are currently available for every state and county in the United States, and for selected international locations. CONCLUSIONS: State and county public health planners can use FRED to explore the effects of possible influenza epidemics in specific geographic regions of interest and to help evaluate the effect of interventions such as vaccination programs and school closure policies. FRED is available under a free open source license in order to contribute to the development of better modeling tools and to encourage open discussion of modeling tools being used to evaluate public health policies. We also welcome participation by other researchers in the further development of FRED.

Dynamic simulation of crime perpetration and reporting to examine community intervention strategies.

OBJECTIVE: To develop a conceptual computational agent-based model (ABM) to explore community-wide versus spatially focused crime reporting interventions to reduce community crime perpetrated by youth. METHOD: Agents within the model represent individual residents and interact on a two-dimensional grid representing an abstract nonempirically grounded community setting. Juvenile agents are assigned initial random probabilities of perpetrating a crime and adults are assigned random probabilities of witnessing and reporting crimes. The agents' behavioral probabilities modify depending on the individual's experience with criminal behavior and punishment, and exposure to community crime interventions. Cost-effectiveness analyses assessed the impact of activating different percentages of adults to increase reporting and reduce community crime activity. Community-wide interventions were compared with spatially focused interventions, in which activated adults were focused in areas of highest crime prevalence. RESULTS: The ABM suggests that both community-wide and spatially focused interventions can be effective in reducing overall offenses, but their relative effectiveness may depend on the intensity and cost of the interventions. Although spatially focused intervention yielded localized reductions in crimes, such interventions were shown to move crime to nearby communities. Community-wide interventions can achieve larger reductions in overall community crime offenses than spatially focused interventions, as long as sufficient resources are available. CONCLUSION: The ABM demonstrates that community-wide and spatially focused crime strategies produce unique intervention dynamics influencing juvenile crime behaviors through the decisions and actions of community adults. It shows how such models might be used to investigate community-supported crime intervention programs by integrating community input and expertise and provides a simulated setting for assessing dimensions of cost comparison and intervention effect sustainability. ABM illustrates how intervention models might be used to investigate community-supported crime intervention programs.

Policies to reduce influenza in the workplace: impact assessments using an agent-based model.

OBJECTIVES: We examined the impact of access to paid sick days (PSDs) and stay-at-home behavior on the influenza attack rate in workplaces. METHODS: We used an agent-based model of Allegheny County, Pennsylvania, with PSD data from the US Bureau of Labor Statistics, standard influenza epidemic parameters, and the probability of staying home when ill. We compared the influenza attack rate among employees resulting from workplace transmission, focusing on the effects of presenteeism (going to work when ill). RESULTS: In a simulated influenza epidemic (R0 = 1.4), the attack rate among employees owing to workplace transmission was 11.54%. A large proportion (72.00%) of this attack rate resulted from exposure to employees engaging in presenteeism. Universal PSDs reduced workplace infections by 5.86%. Providing 1 or 2 "flu days"-allowing employees with influenza to stay home-reduced workplace infections by 25.33% and 39.22%, respectively. CONCLUSIONS: PSDs reduce influenza transmission owing to presenteeism and, hence, the burden of influenza illness in workplaces.

A game dynamic model for vaccine skeptics and vaccine believers: measles as an example.

Widespread avoidance of Measles-Mumps-Rubella vaccination (MMR), with a consequent increase in the incidence of major measles outbreaks, demonstrates that the effectiveness of vaccination programs can be thwarted by the public misperceptions of vaccine risk. By coupling game theory and epidemic models, we examine vaccination choice among populations stratified into two behavioral groups: vaccine skeptics and vaccine believers. The two behavioral groups are assumed to be heterogeneous with respect to their perceptions of vaccine and infection risks. We demonstrate that the pursuit of self-interest among vaccine skeptics often leads to vaccination levels that are suboptimal for a population, even if complete coverage is achieved among vaccine believers. The demand for measles vaccine across populations driven by individual self-interest was found to be more sensitive to the proportion of vaccine skeptics than to the extent to which vaccine skeptics misperceive the risk of vaccine. Furthermore, as the number of vaccine skeptics increases, the probability of infection among vaccine skeptics increases initially, but it decreases once the vaccine skeptics begin receiving the vaccination, if both behavioral groups are vaccinated according to individual self-interest. Our results show that the discrepancy between the coverages of measles vaccine that are driven by self-interest and those driven by population interest becomes larger when the cost of vaccination increases. This research illustrates the importance of public education on vaccine safety and infection risk in order to maintain vaccination levels that are sufficient to maintain herd immunity.

The benefits to all of ensuring equal and timely access to influenza vaccines in poor communities.

When influenza vaccines are in short supply, allocating vaccines equitably among different jurisdictions can be challenging. But justice is not the only reason to ensure that poorer counties have the same access to influenza vaccines as do wealthier ones. Using a detailed computer simulation model of the Washington, D.C., metropolitan region, we found that limiting or delaying vaccination of residents of poorer counties could raise the total number of influenza infections and the number of new infections per day at the peak of an epidemic throughout the region-even in the wealthier counties that had received more timely and abundant vaccine access. Among other underlying reasons, poorer counties tend to have high-density populations and more children and other higher-risk people per household, resulting in more interactions and both increased transmission of influenza and greater risk for worse influenza outcomes. Thus, policy makers across the country, in poor and wealthy areas alike, have an incentive to ensure that poorer residents have equal access to vaccines.

Would school closure for the 2009 H1N1 influenza epidemic have been worth the cost?: a computational simulation of Pennsylvania.

BACKGROUND: During the 2009 H1N1 influenza epidemic, policy makers debated over whether, when, and how long to close schools. While closing schools could have reduced influenza transmission thereby preventing cases, deaths, and health care costs, it may also have incurred substantial costs from increased childcare needs and lost productivity by teachers and other school employees. METHODS: A combination of agent-based and Monte Carlo economic simulation modeling was used to determine the cost-benefit of closing schools (vs. not closing schools) for different durations (range: 1 to 8 weeks) and symptomatic case incidence triggers (range: 1 to 30) for the state of Pennsylvania during the 2009 H1N1 epidemic. Different scenarios varied the basic reproductive rate (R(0)) from 1.2, 1.6, to 2.0 and used case-hospitalization and case-fatality rates from the 2009 epidemic. Additional analyses determined the cost per influenza case averted of implementing school closure. RESULTS: For all scenarios explored, closing schools resulted in substantially higher net costs than not closing schools. For R(0) = 1.2, 1.6, and 2.0 epidemics, closing schools for 8 weeks would have resulted in median net costs of $21.0 billion (95% Range: $8.0 - $45.3 billion). The median cost per influenza case averted would have been $14,185 ($5,423 - $30,565) for R(0) = 1.2, $25,253 ($9,501 - $53,461) for R(0) = 1.6, and $23,483 ($8,870 - $50,926) for R(0) = 2.0. CONCLUSIONS: Our study suggests that closing schools during the 2009 H1N1 epidemic could have resulted in substantial costs to society as the potential costs of lost productivity and childcare could have far outweighed the cost savings in preventing influenza cases.

Vaccination deep into a pandemic wave potential mechanisms for a "third wave" and the impact of vaccination.

BACKGROUND: In December 2009, when the H1N1 influenza pandemic appeared to be subsiding, public health officials and unvaccinated individuals faced the question of whether continued H1N1 immunization was still worthwhile. PURPOSE: To delineate what combinations of possible mechanisms could generate a third pandemic wave and then explore whether vaccinating the population at different rates and times would mitigate the wave. METHODS: As part of ongoing work with the Office of the Assistant Secretary for Preparedness and Response at the USDHHS during the H1N1 influenza pandemic, the University of Pittsburgh Models of Infectious Disease Agent Study team employed an agent-based computer simulation model of the Washington DC metropolitan region to delineate what mechanisms could generate a "third pandemic wave" and explored whether vaccinating the population at different rates and times would mitigate the wave. This model included explicit representations of the region's individuals, school systems, workplaces/commutes, households, and communities. RESULTS: Three mechanisms were identified that could cause a third pandemic wave; substantially increased viral transmissibility from seasonal forcing (changing influenza transmission with changing environmental conditions, i.e., seasons) and progressive viral adaptation; an immune escape variant; and changes in social mixing from holiday school closures. Implementing vaccination for these mechanisms, even during the down-slope of the fall epidemic wave, significantly mitigated the third wave. Scenarios showed the gains from initiating vaccination earlier, increasing the speed of vaccination, and prioritizing population subgroups based on Advisory Committee on Immunization Practices recommendations. CONCLUSIONS: Additional waves in an epidemic can be mitigated by vaccination even when an epidemic appears to be waning.

A computer simulation of vaccine prioritization, allocation, and rationing during the 2009 H1N1 influenza pandemic.

In the fall 2009, the University of Pittsburgh Models of Infectious Disease Agent Study (MIDAS) team employed an agent-based computer simulation model (ABM) of the greater Washington, DC, metropolitan region to assist the Office of the Assistant Secretary of Public Preparedness and Response, Department of Health and Human Services, to address several key questions regarding vaccine allocation during the 2009 H1N1 influenza pandemic, including comparing a vaccinating children (i.e., highest transmitters)-first policy versus the Advisory Committee on Immunization Practices (ACIP)-recommended vaccinating at-risk individuals-first policy. Our study supported adherence to the ACIP (instead of a children-first policy) prioritization recommendations for the H1N1 influenza vaccine when vaccine is in limited supply and that within the ACIP groups, children should receive highest priority.

A computer simulation of employee vaccination to mitigate an influenza epidemic.

BACKGROUND: Better understanding the possible effects of vaccinating employees is important and can help policymakers and businesses plan vaccine distribution and administration logistics, especially with the current H1N1 influenza vaccine in short supply. PURPOSE: This article aims to determine the effects of varying vaccine coverage, compliance, administration rates, prioritization, and timing among employees during an influenza pandemic. METHODS: As part of the H1N1 influenza planning efforts of the Models of Infectious Disease Agent Study network, an agent-based computer simulation model was developed for the Washington DC metropolitan region, encompassing five metropolitan statistical areas. Each simulation run involved introducing 100 infectious individuals to initiate a 1.3 reproductive-rate (R(0)) epidemic, consistent with H1N1 parameters to date. Another set of scenarios represented a R(0)=1.6 epidemic. RESULTS: An unmitigated epidemic resulted in substantial productivity losses (a mean of $112.6 million for a serologic 15% attack rate and $193.8 million for a serologic 25% attack rate), even with the relatively low estimated mortality impact of H1N1. Although vaccinating Advisory Committee on Immunization Practices-defined priority groups resulted in the largest savings, vaccinating all remaining workers captured additional savings and, in fact, reduced healthcare workers' and critical infrastructure workers' chances of infection. Moreover, although employee vaccination compliance affected the epidemic, once 20% compliance was achieved, additional increases in compliance provided less incremental benefit. Even though a vast majority of the workplaces in the DC metropolitan region had fewer than 100 employees, focusing on vaccinating only those in larger firms (> or =100 employees) was just as effective in mitigating the epidemic as trying to vaccinate employees in all workplaces. CONCLUSIONS: Timely vaccination of at least 20% of the large-company workforce can play an important role in epidemic mitigation.

Simulating school closure strategies to mitigate an influenza epidemic.

BACKGROUND: There remains substantial debate over the impact of school closure as a mitigation strategy during an influenza pandemic. The ongoing 2009 H1N1 influenza pandemic has provided an unparalleled opportunity to test interventions with the most up-to-date simulations. METHODS: To assist the Allegheny County Health Department during the 2009 H1N1 influenza pandemic, the University of Pittsburgh Models of Infectious Disease Agents Study group employed an agent-based computer simulation model (ABM) of Allegheny County, Pennsylvania, to explore the effects of various school closure strategies on mitigating influenza epidemics of different reproductive rates (R0). RESULTS: Entire school system closures were not more effective than individual school closures. Any type of school closure may need to be maintained throughout most of the epidemic (ie, at least 8 weeks) to have any significant effect on the overall serologic attack rate. In fact, relatively short school closures (ie, 2 weeks or less) may actually slightly increase the overall attack rate by returning susceptible students back into schools in the middle of the epidemic. Varying the illness threshold at which school closures are triggered did not seem to have substantial impact on the effectiveness of school closures, suggesting that short delays in closing schools should not cause concern. CONCLUSIONS: School closures alone may not be able to quell an epidemic but, when maintained for at least 8 weeks, could delay the epidemic peak for up to a week, providing additional time to implement a second more effective intervention such as vaccination.

On comparing neuronal morphologies with the constrained tree-edit-distance.

The constrained tree-edit-distance provides a computationally practical method for comparing morphologies directly without first extracting distributions of other metrics. The application of the constrained tree-edit-distance to hippocampal dendrites by Heumann and Wittum is reviewed and considered in the context of other applications and potential future uses. The method has been used on neuromuscular projection axons for comparisons of topology as well as on trees for comparing plant architectures with particular parameter sets that may inform future efforts in comparing dendritic morphologies. While clearly practical on a small scale, testing and extrapolation of run-times raise questions as to the practicality of the constrained tree-edit-distance for large-scale data mining projects. However, other more efficient algorithms may make use of it as a gold standard for direct morphological comparison.

High-resolution haplotype block structure in the cattle genome.

BACKGROUND: The Bovine HapMap Consortium has generated assay panels to genotype ~30,000 single nucleotide polymorphisms (SNPs) from 501 animals sampled from 19 worldwide taurine and indicine breeds, plus two outgroup species (Anoa and Water Buffalo). Within the larger set of SNPs we targeted 101 high density regions spanning up to 7.6 Mb with an average density of approximately one SNP per 4 kb, and characterized the linkage disequilibrium (LD) and haplotype block structure within individual breeds and groups of breeds in relation to their geographic origin and use. RESULTS: From the 101 targeted high-density regions on bovine chromosomes 6, 14, and 25, between 57 and 95% of the SNPs were informative in the individual breeds. The regions of high LD extend up to ~100 kb and the size of haplotype blocks ranges between 30 bases and 75 kb (10.3 kb average). On the scale from 1-100 kb the extent of LD and haplotype block structure in cattle has high similarity to humans. The estimation of effective population sizes over the previous 10,000 generations conforms to two main events in cattle history: the initiation of cattle domestication (~12,000 years ago), and the intensification of population isolation and current population bottleneck that breeds have experienced worldwide within the last ~700 years. Haplotype block density correlation, block boundary discordances, and haplotype sharing analyses were consistent in revealing unexpected similarities between some beef and dairy breeds, making them non-differentiable. Clustering techniques permitted grouping of breeds into different clades given their similarities and dissimilarities in genetic structure. CONCLUSION: This work presents the first high-resolution analysis of haplotype block structure in worldwide cattle samples. Several novel results were obtained. First, cattle and human share a high similarity in LD and haplotype block structure on the scale of 1-100 kb. Second, unexpected similarities in haplotype block structure between dairy and beef breeds make them non-differentiable. Finally, our findings suggest that ~30,000 uniformly distributed SNPs would be necessary to construct a complete genome LD map in Bos taurus breeds, and ~580,000 SNPs would be necessary to characterize the haplotype block structure across the complete cattle genome.

SNP-PHAGE--High throughput SNP discovery pipeline.

BACKGROUND: Single nucleotide polymorphisms (SNPs) as defined here are single base sequence changes or short insertion/deletions between or within individuals of a given species. As a result of their abundance and the availability of high throughput analysis technologies SNP markers have begun to replace other traditional markers such as restriction fragment length polymorphisms (RFLPs), amplified fragment length polymorphisms (AFLPs) and simple sequence repeats (SSRs or microsatellite) markers for fine mapping and association studies in several species. For SNP discovery from chromatogram data, several bioinformatics programs have to be combined to generate an analysis pipeline. Results have to be stored in a relational database to facilitate interrogation through queries or to generate data for further analyses such as determination of linkage disequilibrium and identification of common haplotypes. Although these tasks are routinely performed by several groups, an integrated open source SNP discovery pipeline that can be easily adapted by new groups interested in SNP marker development is currently unavailable. RESULTS: We developed SNP-PHAGE (SNP discovery Pipeline with additional features for identification of common haplotypes within a sequence tagged site (Haplotype Analysis) and GenBank (-dbSNP) submissions. This tool was applied for analyzing sequence traces from diverse soybean genotypes to discover over 10,000 SNPs. This package was developed on UNIX/Linux platform, written in Perl and uses a MySQL database. Scripts to generate a user-friendly web interface are also provided with common queries for preliminary data analysis. A machine learning tool developed by this group for increasing the efficiency of SNP discovery is integrated as a part of this package as an optional feature. The SNP-PHAGE package is being made available open source at http://bfgl.anri.barc.usda.gov/ML/snp-phage/. CONCLUSION: SNP-PHAGE provides a bioinformatics solution for high throughput SNP discovery, identification of common haplotypes within an amplicon, and GenBank (dbSNP) submissions. SNP selection and visualization are aided through a user-friendly web interface. This tool is useful for analyzing sequence tagged sites (STSs) of genomic sequences, and this software can serve as a starting point for groups interested in developing SNP markers.

Application of machine learning in SNP discovery.

BACKGROUND: Single nucleotide polymorphisms (SNP) constitute more than 90% of the genetic variation, and hence can account for most trait differences among individuals in a given species. Polymorphism detection software PolyBayes and PolyPhred give high false positive SNP predictions even with stringent parameter values. We developed a machine learning (ML) method to augment PolyBayes to improve its prediction accuracy. ML methods have also been successfully applied to other bioinformatics problems in predicting genes, promoters, transcription factor binding sites and protein structures. RESULTS: The ML program C4.5 was applied to a set of features in order to build a SNP classifier from training data based on human expert decisions (True/False). The training data were 27,275 candidate SNP generated by sequencing 1973 STS (sequence tag sites) (12 Mb) in both directions from 6 diverse homozygous soybean cultivars and PolyBayes analysis. Test data of 18,390 candidate SNP were generated similarly from 1359 additional STS (8 Mb). SNP from both sets were classified by experts. After training the ML classifier, it agreed with the experts on 97.3% of test data compared with 7.8% agreement between PolyBayes and experts. The PolyBayes positive predictive values (PPV) (i.e., fraction of candidate SNP being real) were 7.8% for all predictions and 16.7% for those with 100% posterior probability of being real. Using ML improved the PPV to 84.8%, a 5- to 10-fold increase. While both ML and PolyBayes produced a similar number of true positives, the ML program generated only 249 false positives as compared to 16,955 for PolyBayes. The complexity of the soybean genome may have contributed to high false SNP predictions by PolyBayes and hence results may differ for other genomes. CONCLUSION: A machine learning (ML) method was developed as a supplementary feature to the polymorphism detection software for improving prediction accuracies. The results from this study indicate that a trained ML classifier can significantly reduce human intervention and in this case achieved a 5-10 fold enhanced productivity. The optimized feature set and ML framework can also be applied to all polymorphism discovery software. ML support software is written in Perl and can be easily integrated into an existing SNP discovery pipeline.