The project was developed into three parts: the analysis of p63 isoform in breast tumours; the study of intra-tumour eterogeneicity in metaplastic breast carcinoma; the analysis of oncocytic breast carcinoma. p63 is a sequence-specific DNA-binding factor, homologue of the tumour suppressor and transcription factor p53. The human p63 gene is composed of 15 exons and transcription can occur from two distinct promoters: the transactivating isoforms (TAp63) are generated by a promoter upstream of exon 1, while the alternative promoter located in intron 3 leads to the expression of N-terminal truncated isoforms (ΔNp63). It has been demonstrated that anti-p63 antibodies decorate the majority of squamous cell carcinomas of different organs; moreover tumours with myoepithelial differentiation of the breast show nuclear p63 expression. Two new isoforms have been described with the same sequence as TAp63 and ΔNp63 but lacking exon 4: d4TAp63 and ΔNp73L, respectively. Purpose of the study was to investigate the molecular expression of N-terminal p63 isoforms in benign and malignant breast tissues. In the present study 40 specimens from normal breast, benign lesions, DIN/DCIS, and invasive carcinomas were analyzed by immunohistochemistry and RT-PCR (Reverse Transcriptase-PCR) in order to disclose the patterns of p63 expression. We have observed that the full-length isoforms can be detected in non neoplastic and neoplastic lesions, while the short isoforms are only present in the neoplastic cells of invasive carcinomas. Metaplastic carcinomas of the breast are a heterogeneous group of neoplasms which exhibit varied patterns of metaplasia and differentiation. The existence of such non-modal populations harbouring distinct genetic aberrations may explain the phenotypic diversity observed within a given tumour. Intra-tumour morphological heterogeneity is not uncommon in breast cancer and it can often be appreciated in metaplastic breast carcinomas. Aim of this study was to determine the existence of intra-tumour genetic heterogeneity in metaplastic breast cancers and whether areas with distinct morphological features in a given tumour might be underpinned by distinct patterns of genetic aberrations. 47 cases of metaplastic breast carcinomas were retrieved. Out of the 47 cases, 9 had areas that were of sufficient dimensions to be independently microdissected. Our results indicate that at least some breast cancers are composed of multiple non-modal populations of clonally related cells and provide direct evidence that at least some types of metaplastic breast cancers are composed of multiple non-modal clones harbouring distinct genetic aberrations. Oncocytic tumours represent a distinctive set of lesions with typical granular cytoplasmatic eosinophilia of the neoplastic cells. Only rare example of breast oncocytic carcinomas have been reported in literature and the incidence is probably underestimated. In this study we have analysed 33 cases of oncocytic invasive breast carcinoma of the breast, selected according to morphological and immunohistochemical criteria. These tumours were morphologically classified and studied by immunohistochemistry and aCGH. We have concluded that oncocytic breast carcinoma is a morphologic entity with distinctive ultrastructural and histological features; immunohistochemically is characterized by a luminal profile, it has a frequency of 19.8%, has not distinctive clinical features and, at molecular level, shows a specific constellation of genetic aberration.

No description available.

The advent of the high throughput era has resulted in rapid growth in the availability of large biological datasets. These massive datasets are organized in public or private repositories, encompassing not only DNA but also multiple biomolecules that represent different layers of biological information. The examination and quantification of one such layer are commonly known as "omics," which include the genome, proteome, transcriptome, and metabolome. Currently, it has become commonplace to conduct association analyses between a single omics and a specific phenotype. This practice has significantly enhanced our comprehension of both biological mechanisms and disease, particularly Mendelian disorders. However, the study of a single omics often fails to capture the entirety of variations within a multi-layered mechanism, as well as the interplay between different biological layers, thus not accurately characterizing changes in complex disorders and regulatory systems. Hence, the integration of information from multiple omics has emerged as the prevailing approach, leading to the development of computational tools for conducting multi-omics analyses. These tools are essential for further unraveling the underlying causes of complex diseases. However, the landscape of multi-omics analysis software is highly diverse, offering researchers a wide range of options in terms of purposes, data types, integration methods, and development techniques. This diversity provides tailored pipelines that cater to specific research needs. Yet, it also poses challenges, as the multitude of software options often lacks standardized practices and protocols. Consequently, a universally accepted gold standard is absent, impeding result reproducibility and comparability across different research efforts. To address this issue, we have developed MOUSSE, a novel modular omicsgeneric pipeline for unsupervised data integration. The characteristic of our tool is to use rank-based subject-specific signatures as input to derive from each omics a subject similarity network. This network maintains the informative content of the input data while reducing its size and allows for a graph-based integration of multiple omics. Using the resulting integrated network, the pipeline clusters the subjects andallows researchers to identify biomarkers for each cluster. One aspect that sets MOUSSE apart from other techniques is that it require almost no data preprocessing, making it more robust to noise in the data and more suitable to novel and not yet fully characterized data types. We tested our tool by analyzing ten publicly available benchmark datasets for different types of cancer. Each dataset contained data from three separate omics, namely transcriptome, methylome and miRNAome. The aim of our analysis was two-folded. First, we wanted to demonstrate that MOUSSE was able to identify the different phenotypes of cancers as clusters, second, we aimed to demonstrate that the pipeline was also able to identify biomarkers for each cancer type or progression. Moreover, we compared MOUSSE clustering performance against tenmulti-omics tools tested on the same data, achieving the highest median classification score. Finally, we performed an additional analysis on the biomarkers selected by the pipeline for a selected number of cancer phenotypes, showing that MOUSSE was able to identify the markers underlying disease progression and differential survival rate between cancer phenotypes. Collectively, these results showed that MOUSSE clustering and biomarker identification can be reliable even when the disease is changing. Finally, we successfully compiled and implemented MOUSSE as an R-package. To enhance the pipeline, we incorporated an additional omics dataset. This integration allowed us to optimize the selection of subject-specific signatures and introduced the capability of iteratively running the tool. This means that users can refine their clustering results while reducing the size of candidates, therefore enhancing the overall effectiveness of the software.

In recent years, the field of Statistical Relational Learning (SRL) [1, 2] has produced new, powerful learning methods that are explicitly designed to solve complex problems, such as collective classification, multi-task learning and structured output prediction, which natively handle relational data, noise, and partial information. Statistical-relational methods rely on some First- Order Logic as a general, expressive formal language to encode both the data instances and the relations or constraints between them. The latter encode background knowledge on the problem domain, and are use to restrict or bias the model search space according to the instructions of domain experts. The new tools developed within SRL allow to revisit old computational biology problems in a less ad hoc fashion, and to tackle novel, more complex ones. Motivated by these developments, in this thesis we describe and discuss the application of SRL to three important biological problems, highlighting the advantages, discussing the trade-offs, and pointing out the open problems. In particular, in Chapter 3 we show how to jointly improve the outputs of multiple correlated predictors of protein features by means of a very gen- eral probabilistic-logical consistency layer. The logical layer — based on grounding-specific Markov Logic networks [3] — enforces a set of weighted first-order rules encoding biologically motivated constraints between the pre- dictions. The refiner then improves the raw predictions so that they least violate the constraints. Contrary to canonical methods for the prediction of protein features, which typically take predicted correlated features as in- puts to improve the output post facto, our method can jointly refine all predictions together, with potential gains in overall consistency. In order to showcase our method, we integrate three stand-alone predictors of corre- lated features, namely subcellular localization (Loctree[4]), disulfide bonding state (Disulfind[5]), and metal bonding state (MetalDetector[6]), in a way that takes into account the respective strengths and weaknesses. The ex- perimental results show that the refiner can improve the performance of the underlying predictors by removing rule violations. In addition, the proposed method is fully general, and could in principle be applied to an array of heterogeneous predictions without requiring any change to the underlying software. In Chapter 4 we consider the multi-level protein–protein interaction (PPI) prediction problem. In general, PPIs can be seen as a hierarchical process occurring at three related levels: proteins bind by means of specific domains, which in turn form interfaces through patches of residues. Detailed knowl- edge about which domains and residues are involved in a given interaction has extensive applications to biology, including better understanding of the bind- ing process and more efficient drug/enzyme design. We cast the prediction problem in terms of multi-task learning, with one task per level (proteins, domains and residues), and propose a machine learning method that collec- tively infers the binding state of all object pairs, at all levels, concurrently. Our method is based on Semantic Based Regularization (SBR) [7], a flexible and theoretically sound SRL framework that employs First-Order Logic con- straints to tie the learning tasks together. Contrarily to most current PPI prediction methods, which neither identify which regions of a protein actu- ally instantiate an interaction nor leverage the hierarchy of predictions, our method resolves the prediction problem up to residue level, enforcing con- sistent predictions between the hierarchy levels, and fruitfully exploits the hierarchical nature of the problem. We present numerical results showing that our method substantially outperforms the baseline in several experi- mental settings, indicating that our multi-level formulation can indeed lead to better predictions. Finally, in Chapter 5 we consider the problem of predicting drug-resistant protein mutations through a combination of Inductive Logic Programming [8, 9] and Statistical Relational Learning. In particular, we focus on viral pro- teins: viruses are typically characterized by high mutation rates, which allow them to quickly develop drug-resistant mutations. Mining relevant rules from mutation data can be extremely useful to understand the virus adaptation mechanism and to design drugs that effectively counter potentially resistant mutants. We propose a simple approach for mutant prediction where the in- put consists of mutation data with drug-resistance information, either as sets of mutations conferring resistance to a certain drug, or as sets of mutants with information on their susceptibility to the drug. The algorithm learns a set of relational rules characterizing drug-resistance, and uses them to generate a set of potentially resistant mutants. Learning a weighted combination of rules allows to attach generated mutants with a resistance score as predicted by the statistical relational model and select only the highest scoring ones. Promising results were obtained in generating resistant mutations for both nucleoside and non-nucleoside HIV reverse transcriptase inhibitors. The ap- proach can be generalized quite easily to learning mutants characterized by more complex rules correlating multiple mutations.

Biological regulatory systems are complex due to their role in living organisms in modulating precise responses to changes in internal and external conditions. In this respect, mathematical models have become essential tools to address their complexity for a better understanding of their mechanisms. The vision here, based on integrating experimental and theoretical techniques, provides a systematic means to quantitatively study the characteristics of the interactions that occur in living organisms. The outcome of such an endeavour should provide insights in terms of predictions and quantifications for further investigations in systems and synthetic biology. In this thesis, we establish an integrated modelling framework that can ensure the interaction of experimental biology with the development of quantitative mathematical descriptions of biological systems. To this end, we develop a framework to simulate and analyse biological regulatory systems by integrating different layers of regulatory information. The work herein presents a biological model development workflow in terms of a step by step approach, highlighting challenges and “real life” problems associated with each stage of model development. In the first part, we have focused on applying systems and synthetic biology modelling tools to the phosphate system at the cellular and genetic levels in Escheria coli. Then, we have analysed the interaction mechanisms and the dynamic behaviour of the phosphate starvation response deactivation and evaluated the role of phosphatase activity. We have investigated how the properties of these signalling systems depend on the network structure. Moreover, we have constructed detailed transcriptional regulatory network models and models for promoter design. In the second part, we have designed a multi-level dynamical set up by providing a novel closed loop whole body model of glucose homeostasis coupled with molecular signalling. We have then developed a system embracing the intracellular metabolic level, the cellular level involving the dynamics of the cells, the organ level, and the processes within the whole body. The output of each model directly has been fed with the variables and the parameters of the next aggregated model. This allowed us to observe the metabolic changes that occur at all levels and monitor inter-level communications for Type 2 Diabetes disease.

Unravelling the functioning of the complex processes involved in living systems is a challenging task. Enzymes are involved in almost all of the chemical processes taking place within the cell. They accelerate chemical reactions by forming a complex with the substrate and therefore lowering the reaction activation energy. The characterisation of the enzyme function at the molecular level is a fundamental step, which has several implications and applications in modern biotechnologies. This thesis investigates statistical and relational learning techniques for the characterisation of the enzyme function. The problem is tackled from two sides: the analysis of the enzyme structure and its interactions with other molecules, and the mining of relevant features from the enzyme mutation data. From the first side a pure statistical learning approach is proposed for directly predicting enzyme functional residues. This approach is shown to improve over the current state of the art on several benchmark datasets. The engineered predictors resulting from this investigation are now available to the public of researchers through the CatANalyst web server. Further improvement of the approach is pursued by proposing a supervised clustering technique for collectively predicting all the residues belonging to the same functional site. On the “learning from mutations†side, the focus shifts to the expressivity and interpretability of the learnt models. This thesis proposes novel statistical relational approaches for mining hierarchical features for multiple related tasks. The resistance of viral enzyme mutants to groups of related inhibitors is modelled in a multitask setting. Learnt models are refined on a group or per-task basis at different levels of the hierarchy. The proposed hierarchical approach is shown to provide statistically significant improvements over both single and multitask alternatives. Moreover it has the ability to provide explanation of the models which are themselves hierarchical. A task clustering approach is also proposed for inferring the structure of tasks when it is unknown. Finally, a relational approach is proposed for exploiting the learnt relational rules for generating novel mutations with specific characteristics. This allows to drastically reduce the space of possible mutations to be experimentally assessed. Promising preliminary results are obtained, which highlight the potential of the approach in guiding mutant engineering and in predicting the viral enzyme evolution. These findings can pave the way to further research directions in functional interpretation of biological data by means of machine learning techniques.

Biology is experiencing a paradigm shift since the advent of next generation sequencing technologies. The retrieved data largely exceeds the capability of biologists to investigate all possibilities in the laboratories, hence predictive tools able to guide the research are now a fundamental component of their workflow. Given the central role of proteins in living organisms, in this thesis we focus on their functional analysis and the intrinsic multi-target nature of this task. To this end, we propose different predictive methods, specifically developed to exploit side knowledge among target variables and examples. As a first contribution we face the task of protein-function prediction and more in general of hierarchical-multilabel classification (HMC). We present Ocelot a predictive pipeline for genome-wide protein characterization. It relies on a statistical-relational-learning tool, where the knowledge on the input examples is coded by the combination of multiple kernel matrices, while relations among target variables are expressed as logical constraints. Both, the mislabeling of examples and the infringement of logical rules are penalized by the loss function, but Ocelot do not forces hierarchical consistency. To overcome this limitation, we present AWX, a neural-networks output-layer that guarantees the formal consistency of HMC predictions. The second contribution is VSC, a binary classifier designed to incorporate the concepts of subsampling and locality in the definition of features to be used as the input of a perceptron. A locality-based confidence measure is used to weight the contribution of maximum-margin hyper-planes built by subsampling pairs of examples of opposite class. The rationale is that local methods can be exploited when a multi-target task is expected, but not reflected in the annotation space. The third and last contribution are NES2RA and OneGenE, two approaches for finding candidates to expand known gene regulatory networks. NES2RA adopts variable-subsetting strategies, enabled by volunteer distributed computing, and the PC algorithm to discover candidate causal relationships within each subset of variables. Then, ranking aggregators combine the partial results into a single ranked candidate genes list. OneGenE overcomes the main limitation of NES2RA, i.e. latency, by precomputing candidate expansion lists for each transcript of an organism that are then aggregated on-demand.

Spinal and bulbar muscular atrophy (SBMA) is an X-linked neuromuscular disorder characterized by the progressive dysfunction and loss of lower motor neurons. SBMA is caused by the expansion of a CAG tandem repeat encoding a polyglutamine (polyQ) tract in the androgen receptor (AR) gene. SBMA belongs to the family of polyQ diseases, which includes eight other neurological diseases caused by the same mutation in unrelated genes. PolyQ diseases share common features, such as that polyQ proteins are typically expressed throughout the body, yet they cause specific neuronal loss. It remains to be clarified why specific sub-populations of neurons degenerate in each polyQ disease. The well-known structure and function of AR make SBMA a good model to investigate polyQ disease pathogenesis. Androgen binding to AR results in its nuclear translocation and binding to androgen-responsive elements (AREs) to regulate gene expression. Moreover, AR is highly phosphorylated. Recently, we obtained evidence that phosphorylation of polyQ-AR by cyclin-dependent kinase 2 (CDK2) at serine 96 increases toxicity. This post-translational modification was enriched in neurons. Therefore, we hypothesized that phosphorylation of polyQ-AR at serine 96 modulates its function in response to activation of neuronal activity, a level of regulation altered in SBMA. We carried out a microarray analysis in resting and stimulated neurons in which AR was activated by androgens. Our preliminary results suggest that AR activation drives a differential gene expression program in stimulated neurons. In order to analyze the role of CDK2 and serine 96 phosphorylation in vivo, we deleted one or both CDK2 alleles in SBMA mice. Modulation of CDK2 expression reduced polyQ-AR phosphorylation at serine 96, decreased polyQ-AR accumulation in neurons, and attenuated disease manifestations in SBMA mice. Finally, we carried out an unbiased high-throughput screening of phosphatase and kinase inhibitors. As read-out, we analyzed polyQ-AR nuclear translocation induced by testosterone, in order to identify compounds to lower polyQ-AR toxicity. We isolated 6 phosphatase and 17 kinase inhibitors as modifiers of polyQ-AR nuclear shuttling. Among them, we found two compounds targeting Cdc25, a known activator of CDK2. Cdc25 modulation altered serine 96 phosphorylation, toxicity and transcriptional activity of polyQ-AR in cells. Our results support the idea that Cdc25 represents a potential candidate to develop new therapeutic strategies for SBMA. In summary, our findings show that serine 96 phosphorylation modifies AR physiological functions in neurons and polyQ-AR toxicity in SBMA.

Cell-free expression systems are widely used to synthesize proteins for subsequent further characterization, to manufacture potentially useful commercial end products, and to construct cellular mimics in the laboratory. The first part of the thesis explores the feasibility of preparing two of the commercially available and widely used E. coli-based cell-free expression systems: the PURE System and the S30 Bacterial Extract. The second part focuses on the characterization of in vitro transcription and translation. The third part of the thesis features an example of an application of S30 Bacterial Extract cell-free expression systems i.e. the building of cell-like structures that can work together with engineered bacteria to achieve a predetermined task. Finally, the construction of a microfluidic dialysis device compatible with cell-free synthetic biology projects is presented.

Bacterial Outer Membrane Vesicles (OMVs) are naturally produced by all Gram-negative bacteria and play a key role in their biology and pathogenesis. Over the last few years, OMVs have become an increasingly attractive vaccine platform for three main reasons. First, they contain several Microbe-Associated-Molecular Patterns (MAMPs), crucial for stimulating innate immunity and promoting adaptive immune responses. Second, they can be easily purified from the culture supernatant, thus making their production process inexpensive and scalable. Third, OMVs can be engineered with foreign antigens. However, the OMV platform requires some optimization for a full-blown exploitation. First, OMVs carry a number of endogenous proteins that would be useful to eliminate to avoid possible interference of immune responses toward the vaccine antigens. Second, OMVs carry abundant quantities of lipopolysaccharide (LPS). LPS is a potent stimulator of the immune system, therefore is essential for OMV adjuvaticity, but such adjuvanticity has to be modulated to avoid reactogenicity. In this study, we have addressed the two issues by creating a strain releasing OMVs with a minimal amount of endogenous proteins and containing a detoxified LPS. In particular, we first developed a CRISPR/Cas9-based genome editing tool which allows the inactivation of any “dispensable†gene in two working days. The efficacy and robustness of this tool was validated on 78 “dispensable genes†. Using our CRISPR/Cas9 protocol, an OMV proteome-minimized E. coli strain, named E. coli BL21(DE3)Δ58, deprived of 58 OMV associated proteins was created. We demonstrated that E. coli BL21(DE3)Δ58 had growth kinetics similar to the progenitor strain and featured a remarkable increase in OMV production. Two additional genes involved in the LPS biosynthetic pathway (msbB and pagP) were subsequently inactivated creating E. coli BL21(DE3)Δ60 which released OMVs with a substantially reduced reactogenicity. The exploitation of the two strains in vaccine applications was finally validated. We successfully engineered E. coli BL21(DE3)Δ58 and E. coli BL21(DE3)Δ60 with several different antigens, demonstrating that such antigens compartmentalized with high efficiency in the OMVs. We also demonstrated that the engineered OMVs from E. coli BL21(DE3)Δ58 and E. coli BL21(DE3)Δ60-derived OMVs elicited high antigen-specific antibody and T cell responses.