Biodefense Bioinformatics Blog

Saturday, April 15, 2006

Inferring cytokine protein interaction networks for response to smallpox vaccination

McKinney BA, Crowe JE Jr, Voss HU, Crooke PS, Barney N, Moore JH. Hybrid grammar-based approach to nonlinear dynamical system identification from biological time series. Physical Review E 2006 Feb;73(2 Pt 1):021912 [PubMed]

We introduce a grammar-based hybrid approach to reverse engineering nonlinear ordinary differential equation models from observed time series. This hybrid approach combines a genetic algorithm to search the space of model architectures with a Kalman filter to estimate the model parameters. Domain-specific knowledge is used in a context-free grammar to restrict the search space for the functional form of the target model. We find that the hybrid approach outperforms a pure evolutionary algorithm method, and we observe features in the evolution of the dynamical models that correspond with the emergence of favorable model components. We apply the hybrid method to both artificially generated time series and experimentally observed protein levels from subjects who received the smallpox vaccine. From the observed data, we infer a cytokine protein interaction network for an individual's response to the smallpox vaccine.

Wednesday, March 22, 2006

Open-Source MDR 1.0.0rc

The Dartmouth Computational Genetics Laboratory is pleased to announce the release of version 1.0.0rc1 of our multifactor dimensionality reduction (MDR) software package. This is the first release candidate (rc1) for version 1.0. This new version includes several important new features along with a number of minor tweaks and fixes. We will beta test this version over the next few weeks. Please send us your feedback and suggestions.

Download Open-Source MDR 1.0.0rc1 here.

Major New Features:

1) Attribute construction. At the heart of MDR is a contructive induction algorithm that takes two or more SNPs and creates a new attribute that is inserted into the dataset. The goal here is to change the representation space of the data to make interaction easier to detect using any statistical or computational classifier (e.g. naive Bayes, logistic regression, decision trees, etc.). The new attribute construction tab allows the user to select two or more variables and construct a new single variable that is inserted into the dataset. This new dataset can then be analyzed using MDR or exported to other analysis software packages (e.g. R, SAS, SPSS, Weka). This may also be useful for modeling hierarchical epistasis. For more information about attribute construction please see our new paper in the Journal of Theoretical Biology [PubMed].

2) Interaction dendrograms. This new feature was added to the MDR software to facilitate statistical interpretation of MDR models. This is accomplished using estimates of interaction information (entropy-based measures) to measure the amount of information about the class (e.g. case-control status) that is gained by putting two attributes together using MDR. Here, a distance matrix is estimated using these entropy measures which in turn are used to build a dendrogram using hierarchical cluster analysis. These dendrograms indicate the degree of synergy or redundancy of pairs of attributes. Red lines in the dendrogram indicate synergy while blue lines indicate redundancy or correlation. Interaction dendrograms have been described previously using Cartersian products by Dr. Aleks Jakulin [PDF] and are described in our new paper in the Journal of Theoretical Biology [PubMed].

This work is supported by NIH grant AI59695

Friday, October 21, 2005

Open-Source MDR 0.6.1

The Dartmouth Computational Genetics Laboratory is pleased to announce the release of version 0.6.1 BETA of our open-source Multifactor Dimensionality Reduction (MDR) software package.

Download information can be found here.

New features in MDR 0.6.1 include:

1) Search or wrapper algorithms

We have implemented the first of several different computational search or wrapper algorithms for identifying gene-gene interaction models when there are too many attributes (i.e. SNPs) to exhaustively evaluate. The first wrapper available is a random search. Future additions will include simulated annealing and a genetic algorithm, for example. A deterministic search strategy such as best first will also be added.

2) Additional filter algorithmsWe have added the cross-product odds ratio (OR) to the filter list. This new metric computes the OR for each pairwise combination of levels within an attribute. It also computes the OR comparing each level to all others combined. The largest OR is returned and can be used to filter SNPs for further analysis with MDR. Note that OR of infinity are not returned. Rather, a cell with a zero count is changed to 1 when the OR is computed. This puts the OR on a scale that is easier to plot and easier to compare across other attributes for the purpose of filtering. Additional filter metrics such as information gain will be added soon.

Open-Source MDR 0.5.1

The Dartmouth Computational Genetics Laboratory is pleased to announce the release of version 0.5.1 BETA of our open-source Multifactor Dimensionality Reduction (MDR) software package.

Download information can be found here.

New features in MDR 0.5.1 include:

1) Filter methods

This new feature provides the ability to 'filter' or select from your list of SNPs or other attributes a subset that can be analyzed using MDR. We provide ReliefF and chi-square statistics for filtering. We also provide tools for visualizing the fitness landscape in the form of a line plot, histogram, or raw text. This approach is useful when the number of attributes or variables exceeds the number that is practical for an exhaustive MDR analysis of attribute combinations. For a recent review of the ReliefF algorithm see Robnik-Sikonja M, Kononenko I. 2003. Theoretical and empirical analysis of reliefF and rreliefF. Machine Learning 53:23-69. We are preparing a manuscript reporting results that show ReliefF works much better than chi-square when gene-gene interactions are present. We think this approach will be useful for MDR analysis of genome-wide data.

2) New configuration option

We have included the option to disable the fitness landscape in the configuration tab. Computing and saving the fitness landscape consumes a lot of memory when there are many attribute combinations to evaluate.

3) New statistics

The output for the Best Model tab includes new statistics such as balanced accuracy, chi-square, precision, kappa, and F-measure. Are there others you would like to see? Let us know here!

4) Minor interface changes

We have also implemented some minor interface changes that we think will make MDR easier to use. For example, the Configuration options are now in a tab instead of a new window. We have removed the Raw Results tab due to its redundancy.

5) Source code redesigned

The source code for the new version of MDR has been redsigned for easier use and reuse.

6) Documentation

The documentation we distribute with the MDR software has been updated. The docs describe each of the features and their use.

Plans for the Future:

The next major release of MDR will include several wrapper algorithms such as simulated annealing and genetic algorithms for stochastic searching when an exhaustive search is not possible, as is the case in a genome-wide association studies. We plan to have these new features ready sometime this fall.Remember! The MDR software is still in the beta testing phase. Your feedback is very important to us. Questions, criticisms, and suggestions can be directed to me (jason DOT h DOT moore {at} Dartmouth DOT edu). Feel free to post your feedback at the MDR project on

Saturday, September 10, 2005

Prediction of smallpox outbreak and evaluation of control-measure policy in Japan

A paper by Ohkusa et al. in the April issue of the Journal of Infection and Chemotherapy present a mathematical model for predicting smallpox outbreak in Japan.

Ohkusa Y, Taniguchi K, Okubo I. Prediction of smallpox outbreak and evaluation of control-measure policy in Japan, using a mathematical model. J Infect Chemother. 2005 Apr;11(2):71-80. [PubMed]


Since the September 1 terrorist attacks and moreover, since the anthrax exposure events in 2001 in the United States, bioterrorism attacks seem to be a real threat. Of course, the public health authorities in Japan have started to prepare control measures for such events. We report here our attempts, using a mathematical model, to estimate outbreak size and to examine the most effective measures; comparing ring vaccination (contact tracing, isolation, and vaccination among contacts) and mass vaccination of the susceptible population in the area. The basic framework of the mathematical model follows a model used in previous research. The initial susceptible population is assumed to be 30 million persons. Concerning the important parameters, such as the number of initial-exposure cases, R0 (infectious power, or natural history) and, the starting day of intervention after the initial exposure, we checked the robustness of our conclusions by sensitivity analysis. We found that mass vaccination is preferable to ring vaccination when the values for the initial-exposure cases and R0 are high and when the start of intervention by public health authorities is delayed. In the base-case situation, the mass vaccination strategy needs almost 30 million vaccine doses. On the other hand, though ring vaccination needs fewer doses, it needs fewer than 50,000 doses in the worst-case scenario, that with larger first exposure, higher R0, or later start of public health authority intervention. This mathematical model can measure the prevalence of an infectious disease and can evaluate control measures for it before an outbreak. Especially, it is useful for the planning of the outbreaks of emerging diseases such as severe acute respiratory syndrome (SARS) or for bioterrorism attacks involving such diseases as smallpox. In further research, we will have to take into account the population people vaccinated of for smallpox, who account for about 70% of the total population in Japan.

Statistical challenges in the analysis of two-dimensional difference gel electrophoresis experiments

Fodor et al. present a study published in Bioinformatics on the statistical challenges in the analysis of two-dimensional difference gel electrophoresis experiments using the DeCyderTM software. A case study on the effect of smallpox vaccination is used to compare the results obtained from DeCyder to the results obtained by applying moderated t-tests adjusted for multiple comparisons to DeCyder output data that was additionally normalized.

Fodor IK, Nelson DO, Alegria-Hartman M, Robbins K, Langlois RG, Turteltaub KW, Corzett TH, McCutchen-Maloney SL. Statistical challenges in the analysis of two-dimensional difference gel electrophoresis experiments using DeCyderTM. Bioinformatics. 2005 Aug 9. [PubMed]

Saturday, July 23, 2005

The net of life: reconstructing the microbial phylogenetic network

A new paper by Kunin et al. in Genome Research describes the reconstruction of phylogenetic networks for microbes. Understanding how microbes are related to one another will play an important role in biodefense.

Kunin V, Goldovsky L, Darzentas N, Ouzounis CA. The net of life: reconstructing the microbial phylogenetic network. Genome Res. 2005 Jul;15(7):954-9. [PubMed]


It has previously been suggested that the phylogeny of microbial species might be better described as a network containing vertical and horizontal gene transfer (HGT) events. Yet, all phylogenetic reconstructions so far have presented microbial trees rather than networks. Here, we present a first attempt to reconstruct such an evolutionary network, which we term the "net of life". We use available tree reconstruction methods to infer vertical inheritance, and use an ancestral state inference algorithm to map HGT events on the tree. We also describe a weighting scheme used to estimate the number of genes exchanged between pairs of organisms. We demonstrate that vertical inheritance constitutes the bulk of gene transfer on the tree of life. We term the bulk of horizontal gene flow between tree nodes as "vines", and demonstrate that multiple but mostly tiny vines interconnect the tree. Our results strongly suggest that the HGT network is a scale-free graph, a finding with important implications for genome evolution. We propose that genes might propagate extremely rapidly across microbial species through the HGT network, using certain organisms as hubs.

Thursday, June 23, 2005

MDR 0.4 Released

The Dartmouth Computational Genetics Laboratory is pleased to announce the release of version 0.4 of our open-source multifactor dimensionality reduction (MDR) software package.

MDR 0.4 has been posted to which can be accessed from here.

New features in MDR 0.4 include:

1) Threading to take advantage of multi-processor computers. MDR will now automatically detect if your computer has multiple processors and will parallelize the algorithm accordingly. Thus, if you have two processors with threading turned on, MDR will run 4x faster.

2) Batch/command line mode to allow MDR to be run from scripts. This new feature allows MDR to be run from the command line with a Perl script, for example. This makes it possible to run MDR on a grid or parallel computer for simulation studies.

3) Visualization of the fitness landscape. This new feature plots the training accuracy for every model evaluated by MDR. Line plots or histograms can be selected. A zoom feature permits 'drilling down' on a particular region of the landscape. At a fine resolution, mousing over points reveals the model and the training accuracy of that model.

4) Odds ratios. This statistic and its 95% confidence interval have been added to the MDR output to facilitate an epidemiological interpretation of MDR models.

The next major additions to the MDR software will include computation search or wrapper algorithms for variable or attribute selection when the number of combinations to be evaluated is not computationally feasible. Random, greedy, and stochastic search algorithms will be added. These are necessary for genome-wide association studies. This feature will be available later in the summer.

Is there something you would like to see added to MDR? Request it here.

Note that MDR will be in beta testing for another 2-3 months. Please send us your feedback so we can roll out a polished MDR 1.0 later this summer.

The MDR project is funded by NIH grant AI59694.

Thursday, June 02, 2005

Effects of behavioral changes in a smallpox attack model

A recent paper by Del Valle et al. explores the impact of individual and community behavioral changes in response to an outbreak of smallpox:

Del Valle S, Hethcote H, Hyman JM, Castillo-Chavez C. Effects of behavioral changes in a smallpox attack model. Math Biosci. 2005 Jun;195(2):228-251. [PubMed]


The impact of individual and community behavioral changes in response to an outbreak of a disease with high mortality is often not appreciated. Response strategies to a smallpox bioterrorist attack have focused on interventions such as isolation of infectives, contact tracing, quarantine of contacts, ring vaccination, and mass vaccination. We formulate and analyze a mathematical model in which some individuals lower their daily contact activity rates once an epidemic has been identified in a community. Transmission parameters are estimated from data and an expression is derived for the effective reproduction number. We use computer simulations to analyze the effects of behavior change alone and in combination with other control measures. We demonstrate that the spread of the disease is highly sensitive to how rapidly people reduce their contact activity rates and to the precautions that the population takes to reduce the transmission of the disease. Even gradual and mild behavioral changes can have a dramatic impact in slowing an epidemic. When behavioral changes are combined with other interventions, the epidemic is shortened and the number of smallpox cases is reduced. We conclude that for simulations of a smallpox outbreak to be useful, they must consider the impact of behavioral changes. This is especially true if the model predictions are being used to guide public health policy.

Models for the control of a smallpox outbreak

A new paper by Aldis et al. derives an integral equation model for the control of a smallpox outbreak:

Aldis GK, Roberts MG. An integral equation model for the control of a smallpox outbreak. Math Biosci. 2005 May;195(1):1-22. [PubMed]


An integral equation model of a smallpox epidemic is proposed. The model structures the incidence of infection among the household, the workplace, the wider community and a health-care facility; and incorporates a finite incubation period and plausible infectivity functions. Linearisation of the model is appropriate for small epidemics, and enables analytic expressions to be derived for the basic reproduction number and the size of the epidemic. The effects of control interventions (vaccination, isolation, quarantine and public education) are explored for a smallpox epidemic following an imported case. It is found that the rapid identification and isolation of cases, the quarantine of affected households and a public education campaign to reduce contact would be capable of bringing an epidemic under control. This could be used in conjunction with the vaccination of healthcare workers and contacts. Our results suggest that prior mass vaccination would be an inefficient method of containing an outbreak.

Friday, May 20, 2005

Multifactor Dimensionality Reduction (MDR) Software

An important question in biodefense research is whether DNA sequence variations can predict who is at risk for adverse events following vaccination for bioterroism agents such as smallpox. We have developed and released a multifactor dimensionality reduction (MDR) software package for detecting and characterizing combinations of DNA sequence variations that predict complex clinical endpoints. The Dartmouth Computational Genetics Laboratory (CGL) is pleased to announce the release of MDR 0.3 with many new features. Information about the MDR method can be found here. The open-source MDR software package can be downloaded for free here.

Friday, May 13, 2005

Cost-effectiveness of defending against bioterrorism

A new study by Fowler et al. published in the Annals of Internal Medicine uses descision analytic models to determine the cost-effectiveness of defending against bioterrorism:

Fowler RA, Sanders GD, Bravata DM, Nouri B, Gastwirth JM, Peterson D, Broker AG, Garber AM, Owens DK. Cost-effectiveness of defending against bioterrorism: a comparison of vaccination and antibiotic prophylaxis against anthrax. Ann Intern Med. 2005 Apr 19;142(8):601-10. [PubMed]


BACKGROUND: Weaponized Bacillus anthracis is one of the few biological agents that can cause death and disease in sufficient numbers to devastate an urban setting. OBJECTIVE: To evaluate the cost-effectiveness of strategies for prophylaxis and treatment of an aerosolized B. anthracis bioterror attack. DESIGN: Decision analytic model. DATA SOURCES: We derived probabilities of anthrax exposure, vaccine and treatment characteristics, and their costs and associated clinical outcomes from the medical literature and bioterrorism-preparedness experts. TARGET POPULATION: Persons living and working in a large metropolitan U.S. city. TIME HORIZON: Patient lifetime. PERSPECTIVE: Societal. INTERVENTION: We evaluated 4 postattack strategies: no prophylaxis, vaccination alone, antibiotic prophylaxis alone, or vaccination and antibiotic prophylaxis, as well as preattack vaccination versus no vaccination. OUTCOME MEASURES: Costs, quality-adjusted life-years, life-years, and incremental cost-effectiveness. RESULTS OF BASE-CASE ANALYSIS: If an aerosolized B. anthracis bioweapon attack occurs, postexposure prophylactic vaccination and antibiotic therapy for those potentially exposed is the most effective (0.33 life-year gained per person) and least costly (355 dollars saved per person) strategy, as compared with vaccination alone. At low baseline probabilities of attack and exposure, mass previous vaccination of a metropolitan population is more costly (815 million dollars for a city of 5 million people) and not more effective than no vaccination. RESULTS OF SENSITIVITY ANALYSIS: If prophylactic antibiotics cannot be promptly distributed after exposure, previous vaccination may become cost-effective. LIMITATIONS: The probability of exposure and disease critically depends on the probability and mechanism of bioweapon release. CONCLUSIONS: In the event of an aerosolized B. anthracis bioweapon attack over an unvaccinated metropolitan U.S. population, postattack prophylactic vaccination and antibiotic therapy is the most effective and least expensive strategy.

News and media from Sunnybrook & Women's.