Noam Shental

Adjustment Learning and Relevant Component Analysis

We propose a new learning approach for image retrieval, which we call adjustment learning, and demonstrate its use for face recognition and color matching. Our approach is motivated by a frequently encountered problem, namely, that variability in the original data representation which is not relevant to the task may interfere with retrieval and make it very difficult. Our key observation is that in real applications of image retrieval, data sometimes comes in small chunks - small subsets of images that come from the same (but unknown) class. This is the case, for example, when a query is presented via a short video clip. We call these groups chunklets, and we call the paradigm which uses chunklets for unsupervised learning adjustment learning. Within this paradigm we propose a linear scheme, which we call Relevant Component Analysis; this scheme uses the information in such chunklets to reduce irrelevant variability in the data while amplifying relevant variability. We provide results using our method on two problems: face recognition (using a database publicly available on the web), and visual surveillance (using our own data). In the latter application chunklets are obtained automatically from the data without the need of supervision.

Generalized belief propagation receiver for near-optimal detection of two-dimensional channels with memory

We propose a generalized belief propagation (GBP) receiver for two-dimensional (2D) channels with memory, which is applicative to 2D intersymbol interference (ISI) equalization and multiuser detection (MUD). Our experimental study demonstrates that under non-trivial interference conditions, the performance of this fully tractable GBP receiver is almost identical to the performance of the optimal maximum a-posteriori (MAP) receiver.

Potential role of intratumor bacteria in mediating tumor resistance to the chemotherapeutic drug gemcitabine

In model systems, bacteria present in human pancreatic tumors confer resistance to the anticancer drug gemcitabine. Debugging a cancer therapy Microbes contribute not only to the development of human diseases but also to the response of diseases to treatment. Geller et al. show that certain bacteria express enzymes capable of metabolizing the cancer chemotherapeutic drug gemcitabine into an inactive form. When bacteria were introduced into tumors growing in mice, the tumors became resistant to gemcitabine, an effect that was reversed by antibiotic treatment. Interestingly, a high percentage of human pancreatic ductal adenocarcinomas, a tumor type commonly treated with gemcitabine, contain the culprit bacteria. These correlative results raise the tantalizing possibility that the efficacy of an existing therapy for this lethal cancer might be improved by cotreatment with antibiotics. Science, this issue p. 1156 Growing evidence suggests that microbes can influence the efficacy of cancer therapies. By studying colon cancer models, we found that bacteria can metabolize the chemotherapeutic drug gemcitabine (2′,2′-difluorodeoxycytidine) into its inactive form, 2′,2′-difluorodeoxyuridine. Metabolism was dependent on the expression of a long isoform of the bacterial enzyme cytidine deaminase (CDDL), seen primarily in Gammaproteobacteria. In a colon cancer mouse model, gemcitabine resistance was induced by intratumor Gammaproteobacteria, dependent on bacterial CDDL expression, and abrogated by cotreatment with the antibiotic ciprofloxacin. Gemcitabine is commonly used to treat pancreatic ductal adenocarcinoma (PDAC), and we hypothesized that intratumor bacteria might contribute to drug resistance of these tumors. Consistent with this possibility, we found that of the 113 human PDACs that were tested, 86 (76%) were positive for bacteria, mainly Gammaproteobacteria.

An antibody profile of systemic lupus erythematosus detected by antigen microarray

Patients with systemic lupus erythematosus (SLE) produce antibodies to many different self‐antigens. Here, we investigated antibodies in SLE sera using an antigen microarray containing many hundreds of antigens, mostly self‐antigens. The aim was to detect sets of antibody reactivities characteristic of SLE patients in each of various clinical states – SLE patients with acute lupus nephritis, SLE patients in renal remission, and SLE patients who had never had renal involvement. The analysis produced two novel findings: (i) an SLE antibody profile persists independently of disease activity and despite long‐term clinical remission, and (ii) this SLE antibody profile includes increases in four specific immunoglobulin G (IgG) reactivities to double‐stranded DNA (dsDNA), single‐stranded DNA (ssDNA), Epstein–Barr virus (EBV) and hyaluronic acid; the profile also includes decreases in specific IgM reactivities to myeloperoxidase (MPO), CD99, collagen III, insulin‐like growth factor binding protein 1 (IGFBP1) and cardiolipin. The reactivities together showed high sensitivity (> 93%) and high specificity for SLE (> 88%). A healthy control subject who had the SLE antibody profile was later found to develop clinical SLE. The present study did not detect antibody reactivities that differentiated among the various subgroups of SLE subjects with statistical significance. Thus, SLE is characterized by an enduring antibody profile irrespective of clinical state. The association of SLE with decreased IgM natural autoantibodies suggests that these autoantibodies might enhance resistance to SLE.

Discrete-Input Two-Dimensional Gaussian Channels With Memory: Estimation and Information Rates Via Graphical Models and Statistical Mechanics

Discrete-input two-dimensional (2D) Gaussian channels with memory represent an important class of systems, which appears extensively in communications and storage. In spite of their widespread use, the workings of 2D channels are still very much unknown. In this work, we try to explore their properties from the perspective of estimation theory and information theory. At the heart of our approach is a mapping of a 2D channel to an undirected graphical model, and inferring its a posteriori probabilities (APPs) using generalized belief propagation (GBP). The derived probabilities are shown to be practically accurate, thus enabling optimal maximum a posteriori (MAP) estimation of the transmitted symbols. Also, the Shannon-theoretic information rates are deduced either via the vector-wise Shannon-McMillan-Breiman (SMB) theorem, or via the recently derived symbol-wise Guo-Shamai-Verdu (GSV) theorem. Our approach is also described from the perspective of statistical mechanics, as the graphical model and inference algorithm have their analogues in physics. Our experimental study, based on common channel settings taken from cellular networks and magnetic recording devices, demonstrates that under nontrivial memory conditions, the performance of this fully tractable GBP estimator is almost identical to the performance of the optimal MAP estimator. It also enables a practically accurate simulation-based estimate of the information rate. Rationalization of this excellent performance of GBP in the 2-D Gaussian channel setting is addressed.

Identification of rare alleles and their carriers using compressed se(que)nsing

Identification of rare variants by resequencing is important both for detecting novel variations and for screening individuals for known disease alleles. New technologies enable low-cost resequencing of target regions, although it is still prohibitive to test more than a few individuals. We propose a novel pooling design that enables the recovery of novel or known rare alleles and their carriers in groups of individuals. The method is based on a Compressed Sensing (CS) approach, which is general, simple and efficient. CS allows the use of generic algorithmic tools for simultaneous identification of multiple variants and their carriers. We model the experimental procedure and show via computer simulations that it enables the recovery of rare alleles and their carriers in larger groups than were possible before. Our approach can also be combined with barcoding techniques to provide a feasible solution based on current resequencing costs. For example, when targeting a small enough genomic region (∼100 bp) and using only ∼10 sequencing lanes and ∼10 distinct barcodes per lane, one recovers the identity of 4 rare allele carriers out of a population of over 4000 individuals. We demonstrate the performance of our approach over several publicly available experimental data sets.

Enhancing image and video retrieval: learning via equivalence constraints

The paper is about learning using partial information in the form of equivalence constraints. Equivalence constraints provide relational information about the labels of data points, rather than the labels themselves. Our work is motivated by the observation that in many real life applications partial information about the data can be obtained with very little cost. For example, in video indexing we may want to use the fact that a sequence of faces obtained from successive frames in roughly the same location is likely to contain the same unknown individual. Learning using equivalence constraints is different from learning using labels and poses new technical challenges. In this paper we present three novel methods for clustering and classification, which use equivalence constraints. We provide results of our methods on a distributed image querying system that works on a large facial image database, and on the clustering and retrieval of surveillance data. Our results show that we can significantly improve the performance of image retrieval by taking advantage of such assumptions as temporal continuity in the data. Significant improvement is also obtained by making the users of the system take the role of distributed teachers, which reduces the need for expensive labeling by paid human labor.

High-resolution microbial community reconstruction by integrating short reads from multiple 16S rRNA regions

The emergence of massively parallel sequencing technology has revolutionized microbial profiling, allowing the unprecedented comparison of microbial diversity across time and space in a wide range of host-associated and environmental ecosystems. Although the high-throughput nature of such methods enables the detection of low-frequency bacteria, these advances come at the cost of sequencing read length, limiting the phylogenetic resolution possible by current methods. Here, we present a generic approach for integrating short reads from large genomic regions, thus enabling phylogenetic resolution far exceeding current methods. The approach is based on a mapping to a statistical model that is later solved as a constrained optimization problem. We demonstrate the utility of this method by analyzing human saliva and Drosophila samples, using Illumina single-end sequencing of a 750 bp amplicon of the 16S rRNA gene. Phylogenetic resolution is significantly extended while reducing the number of falsely detected bacteria, as compared with standard single-region Roche 454 Pyrosequencing. Our approach can be seamlessly applied to simultaneous sequencing of multiple genes providing a higher resolution view of the composition and activity of complex microbial communities.

Epstein-Barr virus antibodies mark systemic lupus erythematosus and scleroderma patients negative for anti-DNA

Systemic lupus erythematosus (SLE) is an autoimmune disease that can attack many different body organs; the triggering event is unknown. SLE has been associated with more than 100 different autoantibody reactivities – anti‐dsDNA is prominent. Nevertheless, autoantibodies to dsDNA occur in only two‐thirds of SLE patients. We previously reported the use of an antigen microarray to characterize SLE serology. We now report the results of an expanded study of serology in SLE patients and scleroderma (SSc) patients compared with healthy controls. The analysis validated and extended previous findings: two‐thirds of SLE patients reacted to a large spectrum of self‐molecules that overlapped with their reactivity to dsDNA; moreover, some SLE patients manifested a deficiency of natural IgM autoantibodies. Most significant was the finding that many SLE patients who were negative for autoantibodies to dsDNA manifested abnormal antibody responses to Epstein–Barr virus (EBV): these subjects made IgG antibodies to EBV antigens to which healthy subjects did not respond or they failed to make antibodies to EBV antigens to which healthy subjects did respond. This observation suggests that SLE may be associated with a defective immune response to EBV. The SSc patients shared many of these serological abnormalities with SLE patients, but differed from them in increased IgG autoantibodies to topoisomerase and centromere B; 84% of SLE patients and 58% of SSc patients could be detected by their abnormal antibodies to EBV. Hence an aberrant immune response to a ubiquitous viral infection such as EBV might set the stage for an autoimmune disease.

Delayed development induced by toxicity to the host can be inherited by a bacterial-dependent, transgenerational effect.

Commensal gut bacteria in many species including flies are integral part of their host, and are known to influence its development and homeostasis within generation. Here we report an unexpected impact of host–microbe interactions, which mediates multi-generational, non-Mendelian inheritance of a stress-induced phenotype. We have previously shown that exposure of fly larvae to G418 antibiotic induces transgenerationally heritable phenotypes, including a delay in larval development, gene induction in the gut and morphological changes. We now show that G418 selectively depletes commensal Acetobacter species and that this depletion explains the heritable delay, but not the inheritance of the other phenotypes. Notably, the inheritance of the delay was mediated by a surprising trans-generational effect. Specifically, bacterial removal from F1 embryos did not induce significant delay in F1 larvae, but nonetheless led to a considerable delay in F2. This effect maintains a delay induced by bacterial-independent G418 toxicity to the host. In line with these findings, reintroduction of isolated Acetobacter species prevented the inheritance of the delay. We further show that this prevention is partly mediated by vitamin B2 (Riboflavin) produced by these bacteria; exogenous Riboflavin led to partial prevention and inhibition of Riboflavin synthesis compromised the ability of the bacteria to prevent the inheritance. These results identify host–microbe interactions as a hitherto unrecognized factor capable of mediating non-Mendelian inheritance of a stress-induced phenotype.