Kernel Methods for Tree Structured Data

Machine learning comprises a series of techniques for automatic extraction of meaningful information from large collections of noisy data. In many real world applications, data is naturally represented in structured form. Since traditional methods in machine learning deal with vectorial information, they require an a priori form of preprocessing. Among all the learning techniques for dealing with structured data, kernel methods are recognized to have a strong theoretical background and to be effective approaches. They do not require an explicit vectorial representation of the data in terms of features, but rely on a measure of similarity between any pair of objects of a domain, the kernel function. Designing fast and good kernel functions is a challenging problem. In the case of tree structured data two issues become relevant: kernel for trees should not be sparse and should be fast to compute. The sparsity problem arises when, given a dataset and a kernel function, most structures of the dataset are completely dissimilar to one another. In those cases the classifier has too few information for making correct predictions on unseen data. In fact, it tends to produce a discriminating function behaving as the nearest neighbour rule. Sparsity is likely to arise for some standard tree kernel functions, such as the subtree and subset tree kernel, when they are applied to datasets with node labels belonging to a large domain. A second drawback of using tree kernels is the time complexity required both in learning and classification phases. Such a complexity can sometimes prevents the kernel application in scenarios involving large amount of data. This thesis proposes three contributions for resolving the above issues of kernel for trees. A first contribution aims at creating kernel functions which adapt to the statistical properties of the dataset, thus reducing its sparsity with respect to traditional tree kernel functions. Specifically, we propose to encode the input trees by an algorithm able to project the data onto a lower dimensional space with the property that similar structures are mapped similarly. By building kernel functions on the lower dimensional representation, we are able to perform inexact matchings between different inputs in the original space. A second contribution is the proposal of a novel kernel function based on the convolution kernel framework. Convolution kernel measures the similarity of two objects in terms of the similarities of their subparts. Most convolution kernels are based on counting the number of shared substructures, partially discarding information about their position in the original structure. The kernel function we propose is, instead, especially focused on this aspect. A third contribution is devoted at reducing the computational burden related to the calculation of a kernel function between a tree and a forest of trees, which is a typical operation in the classification phase and, for some algorithms, also in the learning phase. We propose a general methodology applicable to convolution kernels. Moreover, we show an instantiation of our technique when kernels such as the subtree and subset tree kernels are employed. In those cases, Direct Acyclic Graphs can be used to compactly represent shared substructures in different trees, thus reducing the computational burden and storage requirements.

Probabilistic approaches in civil engineering: generation of random fields and structural identification with genetic algorithms

The inherent stochastic character of most of the physical quantities involved in engineering models has led to an always increasing interest for probabilistic analysis. Many approaches to stochastic analysis have been proposed. However, it is widely acknowledged that the only universal method available to solve accurately any kind of stochastic mechanics problem is Monte Carlo Simulation. One of the key parts in the implementation of this technique is the accurate and efficient generation of samples of the random processes and fields involved in the problem at hand. In the present thesis an original method for the simulation of homogeneous, multi-dimensional, multi-variate, non-Gaussian random fields is proposed. The algorithm has proved to be very accurate in matching both the target spectrum and the marginal probability. The computational efficiency and robustness are very good too, even when dealing with strongly non-Gaussian distributions. What is more, the resulting samples posses all the relevant, welldefined and desired properties of “translation fields”, including crossing rates and distributions of extremes. The topic of the second part of the thesis lies in the field of non-destructive parametric structural identification. Its objective is to evaluate the mechanical characteristics of constituent bars in existing truss structures, using static loads and strain measurements. In the cases of missing data and of damages that interest only a small portion of the bar, Genetic Algorithm have proved to be an effective tool to solve the problem.

Biofuels Sustainability Certification Schemes: Challenges, Feasibility and Possible Approaches

The focus of this research is to develop and apply an analytical framework for evaluating the effectiveness and practicability of sustainability certification schemes for biofuels, especially in a developing country’s perspective. The main question that drives the research analysis is “Which are the main elements of and how to develop sustainability certification schemes that would be effective and practicable in certifying the contribution of biofuels in meeting the goals Governments and other stakeholders have set up?”. Biofuels have been identified as a promising tool to reach a variety of goals: climate change protection, energy security, agriculture development, and, especially in developing countries, economic development. Once the goals have been identified, and ambitious mandatory targets for biofuels use agreed at national level, concerns have been raised by the scientific community on the negative externalities that biofuels production and use can have at environment, social and economic level. Therefore certification schemes have been recognized as necessary processes to measure these externalities, and examples of such schemes are in effect, or are in a negotiating phase, both at mandatory and voluntary levels. The research focus has emerged by the concern that the ongoing examples are very demanding in terms of compliance, both for those that are subject to certification and those that have to certify, on the quantity and quality of information to be reported. A certification system, for reasons linked to costs, lack of expertise, inadequate infrastructure, absence of an administrative and legislative support, can represent an intensive burden and can act as a serious impediment for the industrial and agriculture development of developing countries, going against the principle of equity and level playing field. While this research recognizes the importance of comprehensiveness and ambition in designing an important tool for the measurement of sustainability effects of biofuels production and use, it stresses the need to focus on the effectiveness and practicability of this tool in measuring the compliance with the goal. This research that falls under the rationale of the Sustainability Science Program housed at Harvard Kennedy School, has as main objective to close the gap between the research and policy makers worlds in the field of sustainability certification schemes for biofuels.

Different approaches to identify markers for meat quality and production traits in pigs: analysis of the transcriptome in post mortem skeletal muscle and investigation of candidate genes

Due to the growing attention of consumers towards their food, improvement of quality of animal products has become one of the main focus of research. To this aim, the application of modern molecular genetics approaches has been proved extremely useful and effective. This innovative drive includes all livestock species productions, including pork. The Italian pig breeding industry is unique because needs heavy pigs slaughtered at about 160 kg for the production of high quality processed products. For this reason, it requires precise meat quality and carcass characteristics. Two aspects have been considered in this thesis: the application of the transcriptome analysis in post mortem pig muscles as a possible method to evaluate meat quality parameters related to the pre mortem status of the animals, including health, nutrition, welfare, and with potential applications for product traceability (chapters 3 and 4); the study of candidate genes for obesity related traits in order to identify markers associated with fatness in pigs that could be applied to improve carcass quality (chapters 5, 6, and 7). Chapter three addresses the first issue from a methodological point of view. When we considered this issue, it was not obvious that post mortem skeletal muscle could be useful for transcriptomic analysis. Therefore we demonstrated that the quality of RNA extracted from skeletal muscle of pigs sampled at different post mortem intervals (20 minutes, 2 hours, 6 hours, and 24 hours) is good for downstream applications. Degradation occurred starting from 48 h post mortem even if at this time it is still possible to use some RNA products. In the fourth chapter, in order to demonstrate the potential use of RNA obtained up to 24 hours post mortem, we present the results of RNA analysis with the Affymetrix microarray platform that made it possible to assess the level of expression of more of 24000 mRNAs. We did not identify any significant differences between the different post mortem times suggesting that this technique could be applied to retrieve information coming from the transcriptome of skeletal muscle samples not collected just after slaughtering. This study represents the first contribution of this kind applied to pork. In the fifth chapter, we investigated as candidate for fat deposition the TBC1D1 [TBC1 (tre-2/USP6, BUB2, cdc16) gene. This gene is involved in mechanisms regulating energy homeostasis in skeletal muscle and is associated with predisposition to obesity in humans. By resequencing a fragment of the TBC1D1 gene we identified three synonymous mutations localized in exon 2 (g.40A>G, g.151C>T, and g.172T>C) and 2 polymorphisms localized in intron 2 (g.219G>A and g.252G>A). One of these polymorphisms (g.219G>A) was genotyped by high resolution melting (HRM) analysis and PCR-RFLP. Moreover, this gene sequence was mapped by radiation hybrid analysis on porcine chromosome 8. The association study was conducted in 756 performance tested pigs of Italian Large White and Italian Duroc breeds. Significant results were obtained for lean meat content, back fat thickness, visible intermuscular fat and ham weight. In chapter six, a second candidate gene (tribbles homolog 3, TRIB3) is analyzed in a study of association with carcass and meat quality traits. The TRIB3 gene is involved in energy metabolism of skeletal muscle and plays a role as suppressor of adipocyte differentiation. We identified two polymorphisms in the first coding exon of the porcine TRIB3 gene, one is a synonymous SNP (c.132T> C), a second is a missense mutation (c.146C> T, p.P49L). The two polymorphisms appear to be in complete linkage disequilibrium between and within breeds. The in silico analysis of the p.P49L substitution suggests that it might have a functional effect. The association study in about 650 pigs indicates that this marker is associated with back fat thickness in Italian Large White and Italian Duroc breeds in two different experimental designs. This polymorphisms is also associated with lactate content of muscle semimembranosus in Italian Large White pigs. Expression analysis indicated that this gene is transcribed in skeletal muscle and adipose tissue as well as in other tissues. In the seventh chapter, we reported the genotyping results for of 677 SNPs in extreme divergent groups of pigs chosen according to the extreme estimated breeding values for back fat thickness. SNPs were identified by resequencing, literature mining and in silico database mining. analysis, data reported in the literature of 60 candidates genes for obesity. Genotyping was carried out using the GoldenGate (Illumina) platform. Of the analyzed SNPs more that 300 were polymorphic in the genotyped population and had minor allele frequency (MAF) >0.05. Of these SNPs, 65 were associated (P<0.10) with back fat thickness. One of the most significant gene marker was the same TBC1D1 SNPs reported in chapter 5, confirming the role of this gene in fat deposition in pig. These results could be important to better define the pig as a model for human obesity other than for marker assisted selection to improve carcass characteristics.

Analysis and development of ecologically based approaches to coastal defense

Climate-change related impacts, notably coastal erosion, inundation and flooding from sea level rise and storms, will increase in the coming decades enhancing the risks for coastal populations. Further recourse to coastal armoring and other engineered defenses to address risk reduction will exacerbate threats to coastal ecosystems. Alternatively, protection services provided by healthy ecosystems is emerging as a key element in climate adaptation and disaster risk management. I examined two distinct approaches to coastal defense on the base of their ecological and ecosystem conservation values. First, I analyzed the role of coastal ecosystems in providing services for hazard risk reduction. The value in wave attenuation of coral reefs was quantitatively demonstrated using a meta-analysis approach. Results indicate that coral reefs can provide wave attenuation comparable to hard engineering artificial defenses and at lower costs. Conservation and restoration of existing coral reefs are cost-effective management options for disaster risk reduction. Second, I evaluated the possibility to enhance the ecological value of artificial coastal defense structures (CDS) as habitats for marine communities. I documented the suitability of CDS to support native, ecologically relevant, habitat-forming canopy algae exploring the feasibility of enhancing CDS ecological value by promoting the growth of desired species. Juveniles of Cystoseira barbata can be successfully transplanted at both natural and artificial habitats and not affected by lack of surrounding adult algal individuals nor by substratum orientation. Transplantation success was limited by biotic disturbance from macrograzers on CDS compared to natural habitats. Future work should explore the reasons behind the different ecological functioning of artificial and natural habitats unraveling the factors and mechanisms that cause it. The comprehension of the functioning of systems associated with artificial habitats is the key to allow environmental managers to identify proper mitigation options and to forecast the impact of alternative coastal development plans.

Synthesis of Biologically Active Small Molecules: Different Approaches to Drug Design

In the past years, genome biology had disclosed an ever-growing kind of biological targets that emerged as ideal points for therapeutic intervention. Nevertheless, the number of new chemical entities (NCEs) translated into effective therapies employed in the clinic, still not observed. Innovative strategies in drug discovery combined with different approaches to drug design should be searched for bridge this gap. In this context organic synthetic chemistry had to provide for effective strategies to achieve biologically active small molecules to consider not only as potentially drug candidates, but also as chemical tools to dissect biological systems. In this scenario, during my PhD, inspired by the Biology-oriented Synthesis approach, a small library of hybrid molecules endowed with privileged scaffolds, able to block cell cycle and to induce apoptosis and cell differentiation, merged with natural-like cores were synthesized. A synthetic platform which joined a Domino Knoevenagel-Diels Alder reaction with a Suzuki coupling was performed in order to reach the hybrid compounds. These molecules can represent either antitumor lead candidates, or valuable chemical tools to study molecular pathways in cancer cells. The biological profile expressed by some of these derivatives showed a well defined antiproliferative activity on leukemia Bcr-Abl expressing K562 cell lines. A parallel project regarded the rational design and synthesis of minimally structurally hERG blockers with the purpose of enhancing the SAR studies of a previously synthesized collection. A Target-Oriented Synthesis approach was applied. Combining conventional and microwave heating, the desired final compounds were achieved in good yields and reaction rates. The preliminary biological results of the compounds, showed a potent blocking activity. The obtained small set of hERG blockers, was able to gain more insight the minimal structural requirements for hERG liability, which is mandatory to investigate in order to reduce the risk of potential side effects of new drug candidates.

Control of fruit postharvest diseases by thermotherapy: understanding the mechanisms of action by –omics approaches

In recent years the hot water treatment (HW) represents an effective and safe approach for managing postharvest decay. This study reported the effect of an HW (60°C for 60 s and 45°C for 10 min) on brown rot and blue mould respectively. Peaches was found more thermotolerant compared to apple fruit, otherwise Penicillium expansum was more resistant to heat with respect to Monilinia spp. In semi-commercial and commercial trials, the inhibition of brown rot in naturally infected peaches was higher than 78% after 6 days at 0°C and 3 days at 20°C. Moreover, in laboratory trials a 100% disease incidence reduction was obtained by treating artificially infected peaches at 6-12 h after inoculation revealing a curative effect of HW. The expression levels of some genes were evaluated by qRT-PCR. Specifically, the cell wall genes (β-GAL, PL, PG, PME) showed a general decrease of expression level whereas PAL, CHI, HSP70 and ROS-scavenging genes were induced in treated peaches compared to the control ones. Contrarily, HW applied on artificially infected fruit before the inoculum was found to increase brown rot susceptibility. This aspect might be due to an increase of fruit VOCs emission as revealed by PTR-ToF-MS analysis. In addition a microarray experiment was conducted to analyze molecular mechanisms underneath the apple response to heat. Our results showed a largest amount of induced Heat shock proteins (HSPs), Heat shock cognate proteins (HSCs), Heat shock transcription factors (HSTFs) genes found at 1 and 4 hours from the treatment. Those genes required for the thermotolerance process could be involved in induced resistance response. The hypothesis was confirmed by 30% of blue mold disease reduction in artificially inoculated apple after 1 and 4 hours from the treatment. In order to improve peaches quality and disease management during storage, an innovative tool was also used: Da-meter.