Vladimir Jojic

Variation in the human immune system is largely driven by non-heritable influences

There is considerable heterogeneity in immunological parameters between individuals, but its sources are largely unknown. To assess the relative contribution of heritable versus non-heritable factors, we have performed a systems-level analysis of 210 healthy twins between 8 and 82 years of age. We measured 204 different parameters, including cell population frequencies, cytokine responses, and serum proteins, and found that 77% of these are dominated (>50% of variance) and 58% almost completely determined (>80% of variance) by non-heritable influences. In addition, some of these parameters become more variable with age, suggesting the cumulative influence of environmental exposure. Similarly, the serological responses to seasonal influenza vaccination are also determined largely by non-heritable factors, likely due to repeated exposure to different strains. Lastly, in MZ twins discordant for cytomegalovirus infection, more than half of all parameters are affected. These results highlight the largely reactive and adaptive nature of the immune system in healthy individuals.

Systems analysis of sex differences reveals an immunosuppressive role for testosterone in the response to influenza vaccination

Significance There are marked differences between the sexes in their immune response to infections and vaccination, with females often having significantly higher responses. However, the mechanisms underlying these differences are largely not understood. Using a systems immunology approach, we have identified a cluster of genes involved in lipid metabolism and likely modulated by testosterone that correlates with the higher antibody-neutralizing response to influenza vaccination observed in females. Moreover, males with the highest testosterone levels and expression of related gene signatures exhibited the lowest antibody responses to influenza vaccination. This study generates a number of hypotheses on the sex differences observed in the human immune system and their relationship to mechanisms involved in the antibody response to vaccination. Females have generally more robust immune responses than males for reasons that are not well-understood. Here we used a systems analysis to investigate these differences by analyzing the neutralizing antibody response to a trivalent inactivated seasonal influenza vaccine (TIV) and a large number of immune system components, including serum cytokines and chemokines, blood cell subset frequencies, genome-wide gene expression, and cellular responses to diverse in vitro stimuli, in 53 females and 34 males of different ages. We found elevated antibody responses to TIV and expression of inflammatory cytokines in the serum of females compared with males regardless of age. This inflammatory profile correlated with the levels of phosphorylated STAT3 proteins in monocytes but not with the serological response to the vaccine. In contrast, using a machine learning approach, we identified a cluster of genes involved in lipid biosynthesis and previously shown to be up-regulated by testosterone that correlated with poor virus-neutralizing activity in men. Moreover, men with elevated serum testosterone levels and associated gene signatures exhibited the lowest antibody responses to TIV. These results demonstrate a strong association between androgens and genes involved in lipid metabolism, suggesting that these could be important drivers of the differences in immune responses between males and females.

Cytomegalovirus infection enhances the immune response to influenza

Cytomegalovirus infection of young adult humans and mice enhances immune responses to influenza. CMV boosts immune response in the young Cytomegalovirus (CMV) has long been thought of as a sleeper agent—present in a latent form in most people but dangerous when activated in immunosuppressed individuals. Now, Furman et al. look more closely at the effects of CMV infection in young, healthy people. They find that in contrast to aged individuals where CMV infection decreased response to flu vaccine, CMV infection actually enhanced the response to flu vaccine in young adults. This beneficial effect was also seen in mice. These data suggest that latent CMV infection may be beneficial to the host, and provide a possible explanation for the prevalence of CMV infection worldwide. Cytomegalovirus (CMV) is a β-herpesvirus present in a latent form in most people worldwide. In immunosuppressed individuals, CMV can reactivate and cause serious clinical complications, but the effect of the latent state on healthy people remains elusive. We undertook a systems approach to understand the differences between seropositive and negative subjects and measured hundreds of immune system components from blood samples including cytokines and chemokines, immune cell phenotyping, gene expression, ex vivo cell responses to cytokine stimuli, and the antibody response to seasonal influenza vaccination. As expected, we found decreased responses to vaccination and an overall down-regulation of immune components in aged individuals regardless of CMV status. In contrast, CMV-seropositive young adults exhibited enhanced antibody responses to influenza vaccination, increased CD8+ T cell sensitivity, and elevated levels of circulating interferon-γ compared to seronegative individuals. Experiments with young mice infected with murine CMV also showed significant protection from an influenza virus challenge compared with uninfected animals, although this effect declined with time. These data show that CMV and its murine equivalent can have a beneficial effect on the immune response of young, healthy individuals, which may explain the ubiquity of CMV infection in humans and many other species.

Metric Learning from Relative Comparisons by Minimizing Squared Residual

Recent studies [1] -- [5] have suggested using constraints in the form of relative distance comparisons to represent domain knowledge: d(a, b) <; d(c, d) where d(·) is the distance function and a, b, c, d are data objects. Such constraints are readily available in many problems where pairwise constraints are not natural to obtain. In this paper we consider the problem of learning a Mahalanobis distance metric from supervision in the form of relative distance comparisons. We propose a simple, yet effective, algorithm that minimizes a convex objective function corresponding to the sum of squared residuals of constraints. We also extend our model and algorithm to promote sparsity in the learned metric matrix. Experimental results suggest that our method consistently outperforms existing methods in terms of clustering accuracy. Furthermore, the sparsity extension leads to more stable estimation when the dimension is high and only a small amount of supervision is given.

PatchMatch Based Joint View Selection and Depthmap Estimation

We propose a multi-view depthmap estimation approach aimed at adaptively ascertaining the pixel level data associations between a reference image and all the elements of a source image set. Namely, we address the question, what aggregation subset of the source image set should we use to estimate the depth of a particular pixel in the reference image? We pose the problem within a probabilistic framework that jointly models pixel-level view selection and depthmap estimation given the local pairwise image photoconsistency. The corresponding graphical model is solved by EM-based view selection probability inference and PatchMatch-like depth sampling and propagation. Experimental results on standard multi-view benchmarks convey the state-of-the art estimation accuracy afforded by mitigating spurious pixel level data associations. Additionally, experiments on large Internet crowd sourced data demonstrate the robustness of our approach against unstructured and heterogeneous image capture characteristics. Moreover, the linear computational and storage requirements of our formulation, as well as its inherent parallelism, enables an efficient and scalable GPU-based implementation.

Robust multimodal dictionary learning

We propose a robust multimodal dictionary learning method for multimodal images. Joint dictionary learning for both modalities may be impaired by lack of correspondence between image modalities in training data, for example due to areas of low quality in one of the modalities. Dictionaries learned with such non-corresponding data will induce uncertainty about image representation. In this paper, we propose a probabilistic model that accounts for image areas that are poorly corresponding between the image modalities. We cast the problem of learning a dictionary in presence of problematic image patches as a likelihood maximization problem and solve it with a variant of the EM algorithm. Our algorithm iterates identification of poorly corresponding patches and refinements of the dictionary. We tested our method on synthetic and real data. We show improvements in image prediction quality and alignment accuracy when using the method for multimodal image registration.

Learning Microbial Interaction Networks from Metagenomic Count Data.

Many microbes associate with higher eukaryotes and impact their vitality. To engineer microbiomes for host benefit, we must understand the rules of community assembly and maintenance that, in large part, demand an understanding of the direct interactions among community members. Toward this end, we have developed a Poisson-multivariate normal hierarchical model to learn direct interactions from the count-based output of standard metagenomics sequencing experiments. Our model controls for confounding predictors at the Poisson layer and captures direct taxon-taxon interactions at the multivariate normal layer using an ℓ1 penalized precision matrix. We show in a synthetic experiment that our method handily outperforms state-of-the-art methods such as SparCC and the graphical lasso (glasso). In a real in planta perturbation experiment of a nine-member bacterial community, we show our model, but not SparCC or glasso, correctly resolves a direct interaction structure among three community members that associates with Arabidopsis thaliana roots. We conclude that our method provides a structured, accurate, and distributionally reasonable way of modeling correlated count-based random variables and capturing direct interactions among them.

Design of synthetic bacterial communities for predictable plant phenotypes

Specific members of complex microbiota can influence host phenotypes, depending on both the abiotic environment and the presence of other microorganisms. Therefore, it is challenging to define bacterial combinations that have predictable host phenotypic outputs. We demonstrate that plant–bacterium binary-association assays inform the design of small synthetic communities with predictable phenotypes in the host. Specifically, we constructed synthetic communities that modified phosphate accumulation in the shoot and induced phosphate starvation–responsive genes in a predictable fashion. We found that bacterial colonization of the plant is not a predictor of the plant phenotypes we analyzed. Finally, we demonstrated that characterizing a subset of all possible bacterial synthetic communities is sufficient to predict the outcome of untested bacterial consortia. Our results demonstrate that it is possible to infer causal relationships between microbiota membership and host phenotypes and to use these inferences to rationally design novel communities.

Deconvolving molecular signatures of interactions between microbial colonies

Motivation: The interactions between microbial colonies through chemical signaling are not well understood. A microbial colony can use different molecules to inhibit or accelerate the growth of other colonies. A better understanding of the molecules involved in these interactions could lead to advancements in health and medicine. Imaging mass spectrometry (IMS) applied to co-cultured microbial communities aims to capture the spatial characteristics of the colonies’ molecular fingerprints. These data are high-dimensional and require computational analysis methods to interpret. Results: Here, we present a dictionary learning method that deconvolves spectra of different molecules from IMS data. We call this method MOLecular Dictionary Learning (MOLDL). Unlike standard dictionary learning methods which assume Gaussian-distributed data, our method uses the Poisson distribution to capture the count nature of the mass spectrometry data. Also, our method incorporates universally applicable information on common ion types of molecules in MALDI mass spectrometry. This greatly reduces model parameterization and increases deconvolution accuracy by eliminating spurious solutions. Moreover, our method leverages the spatial nature of IMS data by assuming that nearby locations share similar abundances, thus avoiding overfitting to noise. Tests on simulated datasets show that this method has good performance in recovering molecule dictionaries. We also tested our method on real data measured on a microbial community composed of two species. We confirmed through follow-up validation experiments that our method recovered true and complete signatures of molecules. These results indicate that our method can discover molecules in IMS data reliably, and hence can help advance the study of interaction of microbial colonies. Availability and implementation: The code used in this paper is available at: https://github.com/frizfealer/IMS_project. Contact: vjojic@cs.unc.edu Supplementary information: Supplementary data are available at Bioinformatics online.

Semi-coupled dictionary learning for deformation prediction

We propose a coupled dictionary learning method to predict deformation fields based on image appearance. Rather than estimating deformations by standard image registration methods, we investigate how to obtain a basis of the space of deformations. In particular, we explore how image appearance differences with respect to a common atlas image can be used to predict deformations represented by such a basis. We use a coupled dictionary learning method to jointly learn a basis for image appearance differences and their related deformations. Our proposed method is based on local image patches. We evaluate our method on synthetically generated datasets as well as on a structural magnetic resonance brain imaging (MRI) dataset. Our method results in an improved prediction accuracy while reducing the search space compared to nearest neighbor search and demonstrates that learning a deformation basis is feasible.