The current research project mainly aims to investigate the corrosion resistance and bond performance of steel reinforced recycled aggregate concrete incorporating nano-silica under both normal and corrosive environmental conditions. The experimental part includes testing of 180 pull-out specimens prepared from 12 different mixtures. The main parameters studied were the amount of recycled aggregate (RCA) (i.e. 0%, 25%, 50% and 100%), nano silica (1.5% and 3%), steel embedment length as well as steel bar diameter (12 and 20mm). Different levels of corrosion were electrochemically induced by applying impressed voltage technique for 2, 5, 10 and 15 days. The experimental observations mainly focused on the corrosion level in addition to the ultimate bond, failure modes and slips occurred. Experimental results showed that the bond performance between un-corroded steel and recycled aggregate concrete slightly reduced, while a significant degradation was observed after being exposed to corrosive conditions, in comparison to normal concrete. On the other hand, the use of nano silica (NS) showed a reasonable bond enhancement with both normal and RCA concretes under normal conditions. However, much better influence in terms of bond and corrosion resistance was observed under advancing levels of corrosion exposure, reflecting the improvement in corrosion resistance. Therefore, NS was superbly effective in recovering the poor performance in bond for RCA concretes. More efficiency was reported with RCA concretes compared to the conventional concrete. The bond resistance slightly with a small amount of corrosion (almost 2% weight loss), then a significant bond degradation occurs with further corrosion. The influence of specific surface area and amount of nano silica on the performance of concrete with different water/binder (w/b) ratios has been also studied, using 63 different mixtures produced with three different types of colloidal NS having various surface areas and particle sizes. The results showed that the performance of concrete is heavily influenced by changing the surface area of nano silica. Amongst the three used types of nano silica, NS with SSA of 250 m2 /g achieved the highest enhancement rate in terms of compressive strength, water absorption and microstructure analysis, followed by NS with SSA of 500 m2/g, whilst NS with SSA of 51.4 m2 /g was less advantageous for all mixtures. The optimum nano silica ratio in concrete is affected by its particle size as well as water to binder ratio. The feasibility of the impact-echo method for identifying the corrosion was evaluated and compared to the corrosion obtained by mass loss method. The results showed that the impact echo testing can be effectively used to qualitatively detect the damage caused by corrosion in reinforced concrete structures. A significant difference in the dominant frequencies response was observed after exposure to the high and moderate levels of corrosion, whilst no clear trend was observed at the initial stage of corrosion. Artificial neural network models were also developed to predict bond strength for corroded/uncorroded steel bars in concrete using the main influencing parameters (i.e., concrete strength, concrete cover, bar diameter, embedment length and corrosion rate). The developed models were able to predict the bond strength with a high level of accuracy, which was confirmed by conducting a parametric study. / Higher Education Institute of the Libyan Government

The current research project mainly aims to investigate the corrosion resistance and bond performance of steel reinforced recycled aggregate concrete incorporating nano-silica under both normal and corrosive environmental conditions. The experimental part includes testing of 180 pull-out specimens prepared from 12 different mixtures. The main parameters studied were the amount of recycled aggregate (RCA) (i.e. 0%, 25%, 50% and 100%), nano silica (1.5% and 3%), steel embedment length as well as steel bar diameter (12 and 20mm). Different levels of corrosion were electrochemically induced by applying impressed voltage technique for 2, 5, 10 and 15 days. The experimental observations mainly focused on the corrosion level in addition to the ultimate bond, failure modes and slips occurred. Experimental results showed that the bond performance between un-corroded steel and recycled aggregate concrete slightly reduced, while a significant degradation was observed after being exposed to corrosive conditions, in comparison to normal concrete. On the other hand, the use of nano silica (NS) showed a reasonable bond enhancement with both normal and RCA concretes under normal conditions. However, much better influence in terms of bond and corrosion resistance was observed under advancing levels of corrosion exposure, reflecting the improvement in corrosion resistance. Therefore, NS was superbly effective in recovering the poor performance in bond for RCA concretes. More efficiency was reported with RCA concretes compared to the conventional concrete. The bond resistance slightly with a small amount of corrosion (almost 2% weight loss), then a significant bond degradation occurs with further corrosion. The influence of specific surface area and amount of nano silica on the performance of concrete with different water/binder (w/b) ratios has been also studied, using 63 different mixtures produced with three different types of colloidal NS having various surface areas and particle sizes. The results showed that the performance of concrete is heavily influenced by changing the surface area of nano silica. Amongst the three used types of nano silica, NS with SSA of 250 m2 /g achieved the highest enhancement rate in terms of compressive strength, water absorption and microstructure analysis, followed by NS with SSA of 500 m2/g, whilst NS with SSA of 51.4 m2 /g was less advantageous for all mixtures. The optimum nano silica ratio in concrete is affected by its particle size as well as water to binder ratio. The feasibility of the impact-echo method for identifying the corrosion was evaluated and compared to the corrosion obtained by mass loss method. The results showed that the impact-echo testing can be effectively used to qualitatively detect the damage caused by corrosion in reinforced concrete structures. A significant difference in the dominant frequencies response was observed after exposure to the high and moderate levels of corrosion, whilst no clear trend was observed at the initial stage of corrosion. Artificial neural network models were also developed to predict bond strength for corroded/uncorroded steel bars in concrete using the main influencing parameters (i.e., concrete strength, concrete cover, bar diameter, embedment length and corrosion rate). The developed models were able to predict the bond strength with a high level of accuracy, which was confirmed by conducting a parametric study. / Higher Education Institute in the Libyan Government MONE BROS Company in Leeds (UK) for providing recycled aggregates BASF and Akzonobel Companies for providing nano silica NS, Hanson Ltd, UK, for suppling cement

Agricultural industry has started to rely more on data driven approaches to improve productivity and utilize their resources effectively. This thesis project was carried out in collaboration with Ljusgårda AB, it explores plant yield prediction using machine learning models and hyperparameter tweaking. This thesis work is based on data gathered from the company and the plant yield prediction is carried out on two scenarios whereby each scenario is focused on a different time frame of the growth stage. The first scenario predicts yield from day 8 to day 22 of DAT (Day After Transplant), while the second scenario predicts yield from day 1 to day 22 of DAT and three machine learning algorithms Support Vector Regression (SVR), Long Short Time Memory (LSTM) and Artificial Neural Network (ANN) were investigated. Machine learning model’s performances were evaluated using the metrics; Mean Square Error (MSE), Mean Absolute Error (MAE), and r-squared. The evaluation results showed that ANN performed best on MSE and r-squared with dataset 1, while SVR performed best on MAE with dataset 2. Thus, both ANN and SVR meets the objective of this thesis work. The hyperparameter tweaking experiment of the three models further demonstrated the significance of hyperparameter tuning in improving the models and making them more suitable to the available data.

According to the World Health Organization, almost all human population (99%) lives in 117 countries with over 6000 cities, where air pollutant concentration exceeds recommended thresholds. The most common, so-called criteria, air pollutants that affect human lives, are particulate matter (PM) and gas-phase (SO2, CO, NO2, O3 and others). Therefore, many countries or regions worldwide have imposed regulations or interventions to reduce these effects. Whenever an intervention occurs, air quality changes due to changes in ambient factors, such as weather characteristics and human activities. One approach for assessing the effects of interventions or events on air quality is through the use of the Weather Normalized Model (WNM). However, current deterministic models struggle to accurately capture the complex, non-linear relationship between pollutant concentrations and their emission sources. Hence, the primary objective of this thesis is to examine the power of machine learning (ML) and deep learning (DL) techniques to develop and improve WNMs. Subsequently, these enhanced WNMs are employed to assess the impact of events on air quality. Furthermore, these ML/DL-based WNMs can serve as valuable tools for conducting exploratory data analysis (EDA) to uncover the correlations between independent variables (meteorological and temporal features) and air pollutant concentrations within the models.  It has been discovered that DL techniques demonstrated their efficiency and high performance in different fields, such as natural language processing, image processing, biology, and environment. Therefore, several appropriate DL architectures (Long Short-Term Memory - LSTM, Recurrent Neural Network - RNN, Bidirectional Recurrent Neural Network - BIRNN, Convolutional Neural Network - CNN, and Gated Recurrent Unit - GRU) were tested to develop the WNMs presented in Paper I. When comparing these DL architectures and Gradient Boosting Machine (GBM), LSTM-based methods (LSTM, BiRNN) have obtained superior results in developing WNMs. The study also showed that our WNMs (DL-based) could capture the correlations between input variables (meteorological and temporal variables) and five criteria contaminants (SO2, CO, NO2, O3 and PM2.5). This is because the SHapley Additive exPlanations (SHAP) library allowed us to discover the significant factors in DL-based WNMs. Additionally, these WNMs were used to assess the air quality changes during COVID-19 lockdown periods in Ecuador. The existing normalized models operate based on the original units of pollutants and are designed for assessing pollutant concentrations under “average” or consistent weather conditions. Predicting pollution peaks presents an even greater challenge because they often lack discernible patterns. To address this, we enhanced the Weather Normalized Models (WNMs) to boost their performance specifically during daily concentration peak conditions. In the second paper, we accomplished this by developing supervised learning techniques, including Ensemble Deep Learning methods, to distinguish between daily peak and non-peak pollutant concentrations. This approach offers flexibility in categorizing pollutant concentrations as either daily concentration peaks or non-daily concentration peaks. However, it is worth noting that this method may introduce potential bias when selecting non-peak values. In the third paper, WNMs are directly applied to daily concentration peaks to predict and analyse the correlations between meteorological, temporal features and daily concentration peaks of air pollutants.

Traditional methods for locating books and resources in libraries often entail browsing catalogsor manual searching that are time-consuming and inefficient. This thesis investigates thepotential of automated digital services to streamline this process, by utilizing Wi-Fi signal datafor precise indoor localization. Central to this study is the development of a model that employsWi-Fi signal strength (RSSI) and round-trip time (RTT) to estimate the locations of library userswith arm-length accuracy. This thesis aims to enhance the accuracy of location estimation byexploring the complex, nonlinear relationship between Received Signal Strength Indicator(RSSI) and Round-Trip Time (RTT) within signal fingerprints. The model was developed usingan artificial neural network (ANN) to capture the relationship between RSSI and RTT. Besides,this thesis introduces and evaluates the performance of a novel variant of the Particle SwarmOptimization (PSO) algorithm, named Randomized Particle Swarm Optimization (RPSO). Byincorporating randomness into the conventional PSO framework, the RPSO algorithm aims toaddress the limitations of the standard PSO, potentially offering more accurate and reliablelocation estimations. The PSO algorithms, including RPSO, were integrated into the trainingprocess of ANN to optimize the network’s weights and biases through direct optimization, aswell as to enhance the hyperparameters of the ANN’s built-in optimizer. The findings suggestthat optimizing the hyperparameters yields better results than direct optimization of weights andbiases. However, RPSO did not significantly enhance the performance compared to thestandard PSO in this context, indicating the need for further investigation into its application andpotential benefits in complex optimization scenarios.

In this thesis, exergy based SI engine model optimisation (EBSIEMO) is studied and evaluated. A four-stroke bi-fuel spark ignition (SI) engine is modelled for optimisation of engine performance based upon exergy analysis. An artificial neural network (ANN) is used as an emulator to speed up the optimisation processes. Constrained particle swarm optimisation (CPSO) is employed to identify parameters such as equivalence ratio and ignition time for optimising of the engine performance, based upon maximising ¿total availability¿. In the optimisation process, the engine exhaust gases standard emission were applied including brake specific CO (BSCO) and brake specific NOx (BSNOx) as the constraints. The engine model is developed in a two-zone model, while considering the chemical synthesis of fuel, including 10 chemical species. A computer code is developed in MATLAB software to solve the equations for the prediction of temperature and pressure of the mixture in each stage (compression stroke, combustion process and expansion stroke). In addition, Intake and exhaust processes are calculated using an approximation method. This model has the ability to simulate turbulent combustion and compared to computational fluid dynamic (CFD) models it is computationally faster and efficient. The selective outputs are cylinder temperature and pressure, heat transfer, brake work, brake thermal and volumetric efficiency, brake torque, brake power (BP), brake specific fuel consumption (BSFC), brake mean effective pressure (BMEP), concentration of CO2, brake specific CO (BSCO) and brake specific NOx (BSNOx). In this model, the effect of engine speed, equivalence ratio and ignition time on performance parameters using gasoline and CNG fuels are analysed. In addition, the model is validated by experimental data using the results obtained from bi-fuel engine tests. Therefore, this engine model was capable to predict, analyse and useful for optimisation of the engine performance parameters. The exergy based four-stroke bi-fuel (CNG and gasoline) spark ignition (SI) engine model (EBSIEM) here is used for analysis of bi-fuel SI engines. Since, the first law of thermodynamic (the FLT), alone is not able to afford an appropriate comprehension into engine operations. Therefore, this thesis concentrates on the SI engine operation investigation using the developed engine model by the second law of thermodynamic (the SLT) or exergy analysis outlook (exergy based SI engine model (EBSIEM)) In this thesis, an efficient approach is presented for the prediction of total availability, brake specific CO (BSCO), brake specific NOx (BSNOx) and brake torque for bi-fuel engine (CNG and gasoline) using an artificial neural network (ANN) model based on exergy based SI engine (EBSIEM) (ANN-EBSIEM) as an emulator to speed up the optimisation processes. In the other words, the use of a well trained an ANN is ordinarily much faster than mathematical models or conventional simulation programs for prediction. The constrained particle swarm optimisation (CPSO)-EBSIEM (EBSIEMO) was capable of optimising the model parameters for the engine performance. The optimisation results based upon availability analysis (the SLT) due to analysing availability terms, specifically availability destruction (that measured engine irreversibilties) are more regarded with higher priority compared to the FLT analysis. In this thesis, exergy based SI engine model optimisation (EBSIEMO) is studied and evaluated. A four-stroke bi-fuel spark ignition (SI) engine is modelled for optimisation of engine performance based upon exergy analysis. An artificial neural network (ANN) is used as an emulator to speed up the optimisation processes. Constrained particle swarm optimisation (CPSO) is employed to identify parameters such as equivalence ratio and ignition time for optimising of the engine performance, based upon maximising ¿total availability¿. In the optimisation process, the engine exhaust gases standard emission were applied including brake specific CO (BSCO) and brake specific NOx (BSNOx) as the constraints. The engine model is developed in a two-zone model, while considering the chemical synthesis of fuel, including 10 chemical species. A computer code is developed in MATLAB software to solve the equations for the prediction of temperature and pressure of the mixture in each stage (compression stroke, combustion process and expansion stroke). In addition, Intake and exhaust processes are calculated using an approximation method. This model has the ability to simulate turbulent combustion and compared to computational fluid dynamic (CFD) models it is computationally faster and efficient. The selective outputs are cylinder temperature and pressure, heat transfer, brake work, brake thermal and volumetric efficiency, brake torque, brake power (BP), brake specific fuel consumption (BSFC), brake mean effective pressure (BMEP), concentration of CO2, brake specific CO (BSCO) and brake specific NOx (BSNOx). In this model, the effect of engine speed, equivalence ratio and ignition time on performance parameters using gasoline and CNG fuels are analysed. In addition, the model is validated by experimental data using the results obtained from bi-fuel engine tests. Therefore, this engine model was capable to predict, analyse and useful for optimisation of the engine performance parameters. The exergy based four-stroke bi-fuel (CNG and gasoline) spark ignition (SI) engine model (EBSIEM) here is used for analysis of bi-fuel SI engines. Since, the first law of thermodynamic (the FLT), alone is not able to afford an appropriate comprehension into engine operations. Therefore, this thesis concentrates on the SI engine operation investigation using the developed engine model by the second law of thermodynamic (the SLT) or exergy analysis outlook (exergy based SI engine model (EBSIEM)) In this thesis, an efficient approach is presented for the prediction of total availability, brake specific CO (BSCO), brake specific NOx (BSNOx) and brake torque for bi-fuel engine (CNG and gasoline) using an artificial neural network (ANN) model based on exergy based SI engine (EBSIEM) (ANN-EBSIEM) as an emulator to speed up the optimisation processes. In the other words, the use of a well trained an ANN is ordinarily much faster than mathematical models or conventional simulation programs for prediction. The constrained particle swarm optimisation (CPSO)-EBSIEM (EBSIEMO) was capable of optimising the model parameters for the engine performance. The optimisation results based upon availability analysis (the SLT) due to analysing availability terms, specifically availability destruction (that measured engine irreversibilties) are more regarded with higher priority compared to the FLT analysis.

В первом разделе работы рассмотрены факторы, определяющие климатические условия в помещении и методика расчета теплопотерь. Также приведено описание современных способов регулирования температуры в помещении и рассмотрены два реализованных алгоритма для управления температурным режимом в помещении, основанные на математическом анализе и на базе нечеткой логике. Во втором разделе рассматривается тепловая модель здания для традиционной системы регулирования температуры, реализованная в Matlab-Simulink. По рассмотренной модели проведен сравнительный анализ использования различного вида регуляторов и его выбор для реализации интеллектуальной системы управления температурным режимом. В третьем разделе описана реализация алгоритма для предикции температуры на базе нейронных сетей. А также представлено описание реализации аппаратной и программной части интеллектуальной системы. / In the first section of the dissertation, the factors that determine the climatic conditions in the room and the method for calculating heat loss are considered. It also provides a description of modern methods of room temperature control and considers two implemented algorithms for controlling the temperature regime in a room, based on mathematical analysis and on the basis of fuzzy logic. The second section examines a building thermal model for a traditional temperature control system implemented in Matlab-Simulink. Based on the considered model, a comparative analysis of the use of various types of controllers and its choice for the implementation of an intelligent temperature control system is carried out. The third section describes the implementation of an algorithm for predicting temperature based on neural networks. It also provides a description of the implementation of the hardware and software of the intelligent system.

Water contamination is a major source of diseasearound the world. Therefore, the reliable monitoring of harmfulcontamination in water distribution networks requires considerableeffort and attention. It is a vital necessity to possess a reliablemonitoring system in order to detect harmful contamination inwater distribution networks. To measure the potential contamination,a new sensor called an ’electric tongue’ was developedin Link¨opings University. It was created for the purpose ofmeasuring various features of the water reliably. This projecthas developed a supervised machine learning algorithm that usesan artificial neural network for the detection of anomalies in thesystem. The algorithm can detect anomalies with an accuracy ofaround 99.98% based on the data that was available. This wasachieved through a binary classifier, which reconstructs a vectorand compares it to the expected outcome. Despite the limitationsof the problem and the system’s capabilities, binary classificationis a potential solution to this problem. / Vatten kontaminering är en huvudsaklig anledning till sjukdom runtom i världen. Därför är det en avgörande nödvändighet att ha ett tillförlitligt övervakningssystem för att upptäcka skadliga föroreningar i vattendistributionsnät. För att mäta den potentiella föroreningen skapades en ny sensor, den så kallade ”Electric Tongue” vid Linköpings universitet Den skapades i syfte att mäta olika egenskaper i vattnet på ett tillförlitligt sätt. Genom att använda ett artificiellt neuralt nätverk utvecklades en supervised machine learning algoritm för att upptäcka anomalier i systemet. Algoritmen kan upptäcka anomalier med 99.98% säkerhet som baseras på befintliga data. Detta uppnåddes genom att rekonstruera en vektor och jämföra det med det förväntade resultatet genom att använda en binär klassificerare. Trots att det finns begränsningar som orsakats både av problemet men också systemets förmågor, så är binär klassificering en potentiell lösning till detta problem. / Kandidatexjobb i elektroteknik 2022, KTH, Stockholm

Le corps humain contient plus de 3.72X10^13 cellules qui se distinguent par leur morphologie, fonction et état. Leur catalogage en atlas cellulaires c'est entamé il y a plus de 150 ans, avec l'invention des colorants cellulaires en microscopie. Notre connaissance des types cellulaires et leur phénotypes moléculaires nous permet de connaître et prédire leurs fonctions et patrons d'interactions. Ces connaissances sont à la base de la capacité à poser des diagnostics, créer des médicaments et même faire pousser des organes en biologie synthétique. Surprenamment, notre connaissance est loin d'être complète et c'est pourquoi la caractérisation systématique des cellules et l'assemblage des connaissances en atlas cellulaires est nécessaire. Le développement du séquençage à haut débit a révolutionné la biologie des systèmes et ce type de données est parfait pour la construction d'atlas cellulaires entièrement basés sur les données. Un tel atlas cellulaire contiendra une représentation des cellules par des vecteurs de nombres, où chaque vecteur encode le profil moléculaire capturant des informations biologiques de chaque cellule. Chaque expérience de séquençage d'ARN (RNA-Seq) produit des dizaines de milliers de mesures extrêmement riches en information dont l'analyse demeure non-triviale. Des algorithmes de réduction de dimensionnalité, entre autres, permettent d'extraire des données des patrons importants et encoder les échantillons dans des espaces plus interprétables. De cette manière, les cellules similaires sont groupés sur la base d'une multitude de mesures qu'offre le RNA-Seq. Nous avons donc créé un modèle, le Factorized Embedding (FE), qui permet d'organiser les données de séquençage d'ARN de la sorte. Le modèle apprend simultanément deux espaces d'encodage: un pour les échantillons et l'autre pour les gènes. Nous avons observé qu'une fois entraîné, que ce modèle groupe les échantillons sur la base de leur similarité d'expression génique et permet l'interpolation dans l'espace d'encodage et donc une certaine interprétabilité de l'espace d'encodage. Du côté de l'encodage des gènes, nous avons remarqué que les gènes se regroupaient selon leurs patrons de co-expression ainsi que selon des similarité de fonctions, trouvées via des ontologies de gènes (Gene Ontology, GO). Nous avons ensuite exploré les propriétés d'une modification du modèle FE, baptisée le Transcriptome Latent (TLT, de l'anglais The Latent Transcriptome), où l'encodage des gènes est remplacé par une fonction d'encodage de k-mers provenant de données brutes de RNA-Seq. Cette modification du modèle capture dans son espace d'encodage des séquence à la fois de l'information sur la similarité et l'abondance des séquences ADN. L'espace d'encodage a ainsi permis de détecter des anormalités génomiques tels les translocations, ainsi que des mutations spécifiques au patient, rendant cet espace de représentation utile autant pour la visualisation que pour l'analyse de données. Finalement, la dernière itération explorée dans cette thèse, du modèle FE, baptisée cette fois-ci le TCRome, encode des séquences TCR (récepteurs de cellules T) plutôt que des k-mers, venant du séquençage de répertoires immuns (TCR-Seq). Une irrégularité dans la performance du modèle a mené à une analyse des séquences plus approfondie et à la détection de deux sous-types de TCR. Nous avons analysé les répertoires TCR de plus de 1000 individus et rapportons que le répertoire TCR est composé de deux types de TCR ontogéniquement et fonctionellement distincts. Nous avons découvert des patrons distincts dans les abondances de l'un ou l'autre type, changeant en fonction du sexe, l'âge et dans le cadre de maladies telles chez les sujets portant des mutations dans le gène AIRE et dans le cadre de la maladie du greffon contre l'hôte (GVHD). Ces résultats pointent vers la nécessité d'utiliser des données de séquençage multi-modales pour la construction d'atlas cellulaires, c'est à dire en plus des séquence TCR, des données sur l'expression génique ainsi que des caractérisation moléculaires seront probablement utiles, mais leur intégration sera non-triviale. Le modèle FE (et ses modifications) est un bon candidat pour ce type d'encodage, vu sa flexibilité d'architecture et sa résilience aux données manquantes. / The human body contains over 3.72 x 10^13 cells, that distinguish themselves by their morphology, function and state. Their cataloguing into cell atlases has started over 150 years ago, with the invention of cellular stains for microscopy. Our knowledge of cell types and molecular phenotypes allows is to better know and predict their functions and interaction patterns. This knowledge is at the basis of the ability to diagnose disease, create drugs and even grow organs in synthetic biology. Surprisingly, our knowledge is far from complete and this is why a systematic characterization of cells and the assembly of cell atlases is important. The development of high throughput sequencing has revolutionized systems biology and this type of data is perfect for the construction of entirely data-driven cell atlases. Such an atlas will contain a representation of cells by vectors of numbers, where each vector encodes a molecular profile, capturing biological data about each cell. Each sequencing experiment yields tens of thousands of measurements, extremely rich in information, but their analysis remains non-trivial. Dimensionnality reduction algorithms allow to extract from the data important patterns and encode samples into interpretable spaces. This way, similar cells are grouped on the basis of a multitude of measurements that comes from high throughput sequencing. We have created a model, the Factorized Embedding (FE), that allows to organize RNA sequencing (RNA-Seq) data in such a way. The FE model learns simultaneously two encoding spaces: one for samples and one for genes. We have found that the model groups samples on the basis of similar gene expression and allows for smooth interpolation in the encoding space and thus some manner of interpretability. As for the gene encoding space, we observed that gene coordinates were grouped according to co-expression patterns as well as similarity in function, found via gene ontology (GO). We then explored a modification of the FE model, names The Latent Transcriptome (TLT), where the gene encoding function is replaced by a function encoding k-mers, calculated from raw RNA-Seq data. This modification of the model captured in the k-mer encoding space both sequence similarity and sequence abundance. The encoding space allowed for the detection of genomic abnormalities such as translocations, as well as patient-specific mutations, making the encoding space useful for both visualisation and data analysis. Finally, the last iteration of the FE model that we explored, called TCRome, encodes amino-acid TCR sequences rather than k-mers. An irregularity in the model's performance led us to discover two TCR subtypes, entirely based on their sequence. We have thus analyzed TCR repertoires of over 1000 individuals and report that the TCR repertoire is composed of two ontogenically and functionally distinct types. We have discovered distinct pattens in the abundances of each of the sub-types, changing with age, sex and in the context of some diseases such as in individuals carrying a mutated AIRE gene and in graft versus host disease (GVHD). Collectively, these results point towards the necessity to use multi-modal sequencing data for the construction of cell atlases, namely gene expression data, TCR sequencing data and possibly various molecular characterizations. The integration of all this data will however be non-trivial. The FE model (and its modifications) is a good candidate for this type of data organisation, namely because of its flexibility in architecture and resilience to missing data.

No description available.