Learning with Kernels on Graphs: DAG-based kernels, data streams and RNA function prediction.

In many application domains data can be naturally represented as graphs. When the application of analytical solutions for a given problem is unfeasible, machine learning techniques could be a viable way to solve the problem. Classical machine learning techniques are defined for data represented in a vectorial form. Recently some of them have been extended to deal directly with structured data. Among those techniques, kernel methods have shown promising results both from the computational complexity and the predictive performance point of view. Kernel methods allow to avoid an explicit mapping in a vectorial form relying on kernel functions, which informally are functions calculating a similarity measure between two entities. However, the definition of good kernels for graphs is a challenging problem because of the difficulty to find a good tradeoff between computational complexity and expressiveness. Another problem we face is learning on data streams, where a potentially unbounded sequence of data is generated by some sources. There are three main contributions in this thesis. The first contribution is the definition of a new family of kernels for graphs based on Directed Acyclic Graphs (DAGs). We analyzed two kernels from this family, achieving state-of-the-art results from both the computational and the classification point of view on real-world datasets. The second contribution consists in making the application of learning algorithms for streams of graphs feasible. Moreover,we defined a principled way for the memory management. The third contribution is the application of machine learning techniques for structured data to non-coding RNA function prediction. In this setting, the secondary structure is thought to carry relevant information. However, existing methods considering the secondary structure have prohibitively high computational complexity. We propose to apply kernel methods on this domain, obtaining state-of-the-art results.

A non-coding genomic duplication at the HMX1 locus is associated with crop ears in highland cattle.

Highland cattle with congenital crop ears have notches of variable size on the tips of both ears. In some cases, cartilage deformation can be seen and occasionally the external ears are shortened. We collected 40 cases and 80 controls across Switzerland. Pedigree data analysis confirmed a monogenic autosomal dominant mode of inheritance with variable expressivity. All affected animals could be traced back to a single common ancestor. A genome-wide association study was performed and the causative mutation was mapped to a 4 Mb interval on bovine chromosome 6. The H6 family homeobox 1 (HMX1) gene was selected as a positional and functional candidate gene. By whole genome re-sequencing of an affected Highland cattle, we detected 6 non-synonymous coding sequence variants and two variants in an ultra-conserved element at the HMX1 locus with respect to the reference genome. Of these 8 variants, only a non-coding 76 bp genomic duplication (g.106720058_106720133dup) located in the conserved region was perfectly associated with crop ears. The identified copy number variation probably results in HMX1 misregulation and possible gain-of-function. Our findings confirm the role of HMX1 during the development of the external ear. As it is sometimes difficult to phenotypically diagnose Highland cattle with slight ear notches, genetic testing can now be used to improve selection against this undesired trait.

Approximate decoding approaches for network coded correlated data

This paper considers a framework where data from correlated sources are transmitted with the help of network coding in ad hoc network topologies. The correlated data are encoded independently at sensors and network coding is employed in the intermediate nodes in order to improve the data delivery performance. In such settings, we focus on the problem of reconstructing the sources at decoder when perfect decoding is not possible due to losses or bandwidth variations. We show that the source data similarity can be used at decoder to permit decoding based on a novel and simple approximate decoding scheme. We analyze the influence of the network coding parameters and in particular the size of finite coding fields on the decoding performance. We further determine the optimal field size that maximizes the expected decoding performance as a trade-off between information loss incurred by limiting the resolution of the source data and the error probability in the reconstructed data. Moreover, we show that the performance of the approximate decoding improves when the accuracy of the source model increases even with simple approximate decoding techniques. We provide illustrative examples showing how the proposed algorithm can be deployed in sensor networks and distributed imaging applications.

Decoding Delay Minimization in Inter-Session Network Coding

Intra-session network coding has been shown to offer significant gains in terms of achievable throughput and delay in settings where one source multicasts data to several clients. In this paper, we consider a more general scenario where multiple sources transmit data to sets of clients over a wireline overlay network. We propose a novel framework for efficient rate allocation in networks where intermediate network nodes have the opportunity to combine packets from different sources using randomized network coding. We formulate the problem as the minimization of the average decoding delay in the client population and solve it with a gradient-based stochastic algorithm. Our optimized inter-session network coding solution is evaluated in different network topologies and is compared with basic intra-session network coding solutions. Our results show the benefits of proper coding decisions and effective rate allocation for lowering the decoding delay when the network is used by concurrent multicast sessions.

RC-NDN: Raptor Codes Enabled Named Data Networking

Information-centric networking (ICN) has been proposed to cope with the drawbacks of the Internet Protocol, namely scalability and security. The majority of research efforts in ICN have focused on routing and caching in wired networks, while little attention has been paid to optimizing the communication and caching efficiency in wireless networks. In this work, we study the application of Raptor codes to Named Data Networking (NDN), which is a popular ICN architecture, in order to minimize the number of transmitted messages and accelerate content retrieval times. We propose RC-NDN, which is a NDN compatible Raptor codes architecture. In contrast to other coding-based NDN solutions that employ network codes, RC-NDN considers security architectures inherent to NDN. Moreover, different from existing network coding based solutions for NDN, RC-NDN does not require significant computational resources, which renders it appropriate for low cost networks. We evaluate RC-NDN in mobile scenarios with high mobility. Evaluations show that RC-NDN outperforms the original NDN significantly. RC-NDN is particularly efficient in dense environments, where retrieval times can be reduced by 83% and the number of Data transmissions by 84.5% compared to NDN.

Temporal trends in the systemic inflammatory response syndrome, sepsis, and medical coding of sepsis.

BACKGROUND Recent reports using administrative claims data suggest the incidence of community- and hospital-onset sepsis is increasing. Whether this reflects changing epidemiology, more effective diagnostic methods, or changes in physician documentation and medical coding practices is unclear. METHODS We performed a temporal-trend study from 2008 to 2012 using administrative claims data and patient-level clinical data of adult patients admitted to Barnes-Jewish Hospital in St. Louis, Missouri. Temporal-trend and annual percent change were estimated using regression models with autoregressive integrated moving average errors. RESULTS We analyzed 62,261 inpatient admissions during the 5-year study period. 'Any SIRS' (i.e., SIRS on a single calendar day during the hospitalization) and 'multi-day SIRS' (i.e., SIRS on 3 or more calendar days), which both use patient-level data, and medical coding for sepsis (i.e., ICD-9-CM discharge diagnosis codes 995.91, 995.92, or 785.52) were present in 35.3 %, 17.3 %, and 3.3 % of admissions, respectively. The incidence of admissions coded for sepsis increased 9.7 % (95 % CI: 6.1, 13.4) per year, while the patient data-defined events of 'any SIRS' decreased by 1.8 % (95 % CI: -3.2, -0.5) and 'multi-day SIRS' did not change significantly over the study period. Clinically-defined sepsis (defined as SIRS plus bacteremia) and severe sepsis (defined as SIRS plus hypotension and bacteremia) decreased at statistically significant rates of 5.7 % (95 % CI: -9.0, -2.4) and 8.6 % (95 % CI: -4.4, -12.6) annually. All-cause mortality, SIRS mortality, and SIRS and clinically-defined sepsis case fatality did not change significantly during the study period. Sepsis mortality, based on ICD-9-CM codes, however, increased by 8.8 % (95 % CI: 1.9, 16.2) annually. CONCLUSIONS The incidence of sepsis, defined by ICD-9-CM codes, and sepsis mortality increased steadily without a concomitant increase in SIRS or clinically-defined sepsis. Our results highlight the need to develop strategies to integrate clinical patient-level data with administrative data to draw more accurate conclusions about the epidemiology of sepsis.

NetCodCCN: a Network Coding approach for Content-Centric Networks

Content-Centric Networking (CCN) naturally supports multi-path communication, as it allows the simultaneous use of multiple interfaces (e.g. LTE and WiFi). When multiple sources and multiple clients are considered, the optimal set of distribution trees should be determined in order to optimally use all the available interfaces. This is not a trivial task, as it is a computationally intense procedure that should be done centrally. The need for central coordination can be removed by employing network coding, which also offers improved resiliency to errors and large throughput gains. In this paper, we propose NetCodCCN, a protocol for integrating network coding in CCN. In comparison to previous works proposing to enable network coding in CCN, NetCodCCN permit Interest aggregation and Interest pipelining, which reduce the data retrieval times. The experimental evaluation shows that the proposed protocol leads to significant improvements in terms of content retrieval delay compared to the original CCN. Our results demonstrate that the use of network coding adds robustness to losses and permits to exploit more efficiently the available network resources. The performance gains are verified for content retrieval in various network scenarios.

An intergenic non-coding RNA targets and regulates the ribosome in H. volcanii

As translation is the final step in gene expression it is particularly important to understand the processes involved in translation regulation. It was shown in the last years that a class of RNA, the non-protein-coding RNAs (ncRNAs), is involved in regulation of gene expression via various mechanisms (e.g. gene silencing by microRNAs). Almost all of these ncRNA discovered so far target the mRNA in order to modulate protein biosynthesis, this is rather unexpected considering the crucial role of the ribosome during gene expression. However, recent data from our laboratory showed that there is a new class of ncRNAs, which target the ribosome itself [Gebetsberger et al., 2012/ Pircher et al, 2014]. These so called ribosome-associated ncRNAs (rancRNAs) have an impact on translation regulation, mainly by interfering / modulating the rate of protein biosynthesis. The main goal of this project is to identify and describe novel potential regulatory rancRNAs in H. volcanii with the focus on intergenic candidates. Northern blot analyses already revealed interactions with the ribosome and showed differential expression of rancRNAs during different growth phases or under specific stress conditions. To investigate the biological relevance of these rancRNAs, knock-outs were generated in H. volcanii which were used for phenotypic characterization studies. The rancRNA s194 showed association with the 50S ribosomal subunit in vitro and in vivo and was capable of inhibiting peptide bond formation and seems to inhibit translation in vitro. These preliminary data for the rancRNA s194 make it an interesting candidate for further functional studies to identify the molecular mechanisms by which rancRNAs can modulate protein biosynthesis. Characterization of further rancRNA candidates are also underway.

Evaluation of therapist and client language in Motivational Interviewing (MI) sessions: A secondary analysis of data from the Southern Methodist Alcohol Research Trial (SMART) study

In the United States, “binge” drinking among college students is an emerging public health concern due to the significant physical and psychological effects on young adults. The focus is on identifying interventions that can help decrease high-risk drinking behavior among this group of drinkers. One such intervention is Motivational interviewing (MI), a client-centered therapy that aims at resolving client ambivalence by developing discrepancy and engaging the client in change talk. Of late, there is a growing interest in determining the active ingredients that influence the alliance between the therapist and the client. This study is a secondary analysis of the data obtained from the Southern Methodist Alcohol Research Trial (SMART) project, a dismantling trial of MI and feedback among heavy drinking college students. The present project examines the relationship between therapist and client language in MI sessions on a sample of “binge” drinking college students. Of the 126 SMART tapes, 30 tapes (‘MI with feedback’ group = 15, ‘MI only’ group = 15) were randomly selected for this study. MISC 2.1, a mutually exclusive and exhaustive coding system, was used to code the audio/videotaped MI sessions. Therapist and client language were analyzed for communication characteristics. Overall, therapists adopted a MI consistent style and clients were found to engage in change talk. Counselor acceptance, empathy, spirit, and complex reflections were all significantly related to client change talk (p-values ranged from 0.001 to 0.047). Additionally, therapist ‘advice without permission’ and MI Inconsistent therapist behaviors were strongly correlated with client sustain talk (p-values ranged from 0.006 to 0.048). Simple linear regression models showed a significant correlation between MI consistent (MICO) therapist language (independent variable) and change talk (dependent variable) and MI inconsistent (MIIN) therapist language (independent variable) and sustain talk (dependent variable). The study has several limitations such as small sample size, self-selection bias, poor inter-rater reliability for the global scales and the lack of a temporal measure of therapist and client language. Future studies might consider a larger sample size to obtain more statistical power. In addition the correlation between therapist language, client language and drinking outcome needs to be explored.^

X-ray fluorescence scanning of discrete samples on the Schwalbenberg II loess-paleosol sequence, raw data, magnetic susceptibility, organic carbon and U-ratio as grainsize index

The Schwalbenberg II loess-paleosol sequence (LPS) denotes a key site for Marine Isotope Stage (MIS 3) in Western Europe owing to eight succeeding cambisols, which primarily constitute the Ahrgau Subformation. Therefore, this LPS qualifies as a test candidate for the potential of temporal high-resolution geochemical data obtained X-ray fluorescence (XRF) scanning of discrete samplesproviding a fast and non-destructive tool for determining the element composition. The geochemical data is first contextualized to existing proxy data such as magnetic susceptibility (MS) and organic carbon (Corg) and then aggregated to element log ratios characteristic for weathering intensity [LOG (Ca/Sr), LOG (Rb/Sr), LOG (Ba/Sr), LOG (Rb/K)] and dust provenance [LOG (Ti/Zr), LOG (Ti/Al), LOG (Si/Al)]. Generally, an interpretation of rock magnetic particles is challenged in western Europe, where not only magnetic enhancement but also depletion plays a role. Our data indicates leaching and top-soil erosion induced MS depletion at the Schwalbenberg II LPS. Besides weathering, LOG (Ca/Sr) is susceptible for secondary calcification. Thus, also LOG (Rb/Sr) and LOG (Ba/Sr) are shown to be influenced by calcification dynamics. Consequently, LOG (Rb/K) seems to be the most suitable weathering index identifying the Sinzig Soils S1 and S2 as the most pronounced paleosols for this site. Sinzig Soil S3 is enclosed by gelic gleysols and in contrast to S1 and S2 only initially weathered pointing to colder climate conditions. Also the Remagen Soils are characterized by subtle to moderate positive excursions in the weathering indices. Comparing the Schwalbenberg II LPS with the nearby Eifel Lake Sediment Archive (ELSA) and other more distant German, Austrian and Czech LPS while discussing time and climate as limiting factors for pedogenesis, we suggest that the lithologically determined paleosols are in-situ soil formations. The provenance indices document a Zr-enrichment at the transition from the Ahrgau to the Hesbaye Subformation. This is explained by a conceptual model incorporating multiple sediment recycling and sorting effects in eolian and fluvial domains.

Plant protein-coding gene families: emerging bioinformatics approaches

Protein-coding gene families are sets of similar genes with a shared evolutionary origin and, generally, with similar biological functions. In plants, the size and role of gene families has been only partially addressed. However, suitable bioinformatics tools are being developed to cluster the enormous number of sequences currently available in databases. Specifically, comparative genomic databases promise to become powerful tools for gene family annotation in plant clades. In this review, I evaluate the data retrieved from various gene family databases, the ease with which they can be extracted and how useful the extracted information is.

Multi-Resolution Texture Coding for Multi-Resolution 3D Meshes

We present an innovative system to encode and transmit textured multi-resolution 3D meshes in a progressive way, with no need to send several texture images, one for each mesh LOD (Level Of Detail). All texture LODs are created from the finest one (associated to the finest mesh), but can be re- constructed progressively from the coarsest thanks to refinement images calculated in the encoding process, and transmitted only if needed. This allows us to adjust the LOD/quality of both 3D mesh and texture according to the rendering power of the device that will display them, and to the network capacity. Additionally, we achieve big savings in data transmission by avoiding altogether texture coordinates, which are generated automatically thanks to an unwrapping system agreed upon by both encoder and decoder.