Predictive Modeling Tools Research Articles

Development of a Formula for Calculating the Probability of the Spread of African Swine Fever

Robust epidemiological knowledge and predictive modelling tools are needed to address challenging objectives forecasting epidemics. Often, multiple modelling approaches can be used during an epidemic to support effective decision making in a timely manner. A formula has been developed for calculating the probability of spreading african swine fever from infected animals of a threatened pig farm to other farms. The formula allows you to calculate the probability of an outbreak of infection for pigs of each of the pig farms located in the analyzed area. This takes into account the possible ways of spreading the pathogen in the presence of mechanisms and factors of transmission of african swine fever in real time.

How to responsibly deploy a predictive modelling dashboard for study advisors? A use case illustrating various stakeholder perspectives

Most higher education institutions employ study advisors to support their students. To adequately perform their task, study advisors have access to study information about their students. Using AI techniques to analyze that information and to predict if a student might be at risk of study delay could be a valuable tool in study advisors' practice. In this paper, we present a use case of how such a tool was developed (in the form of a dashboard) and which steps and considerations played a role in the responsible deployment of the tool. Three aspects are described: first, we present the timeline of the case study and zoom in on how the macro-level of the institution (where the groundwork is laid to facilitate AI-systems in education) and the micro-level of the implementation of the system influenced each other. Second, we describe which stakeholders were involved and what their ethical considerations were concerning data management, algorithms, and pedagogy. Third, we describe the initial evaluation of the dashboard in terms of study advisors’ experiences and provide suggestions on how to stimulate the responsible and useful implementation of a predictive modelling tool.

Comparative study of statistical computational approaches to investigate the degraded compressive strength of concrete under the freeze-thaw effect

The primary cause of frost damage and early failure in concrete structures is the repeated freezing and thaw cycles (FTC). The evolution of internal fractures and scaling at the surface of concrete can be used to assess the impact of frost damage. Surface scaling mechanisms and interior frost damage are contingent upon numerous environmental factors, including the rate of freezing, low temperature, and duration of the freezing point. However, evaluating the amount of strength loss of concrete material is a challenging factor to consider. This work explores the application of predictive modelling tools, including random forest (RF), multilayer perceptron (MLP), decision tree (DT), and bagging, to evaluate the degraded compressive strength (D-CS) of concrete. The models use five input variables: initial compressive strength (I-CS), water-to-cement ratio (W/C), FTC, minimum temperature (Tmin), and maximum temperature (Tmax), with D-CS as the output. Model performance was compared using mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) metrics, with the bagging model achieving the highest predictive performance at 92 %. A 10-fold cross-validation approach validated the model's accuracy. The influence of each variable was assessed using SHapley Additive exPlanations (SHAP) analysis. Importantly, a user-friendly Graphical User Interface (GUI) was developed based on the models, making it easy for researchers and professionals to make predictions and thereby increasing the practicality and accessibility of this research. This study aids the research community in selecting appropriate models to forecast the strength of various concrete types, thereby enhancing the practical application of this research.

Natural Language Processing Tools for Predictive Modeling of Advanced Trends in Formal Ontologies in Biomedical Sciences

Natural Language Processing Tools for Predictive Modeling of Advanced Trends in Formal Ontologies in Biomedical Sciences

Neural Network Models for Ionospheric Electron Density Prediction at a Fixed Altitude Using Neural Architecture Search

AbstractSpecification and forecast of ionospheric parameters, such as ionospheric electron density (Ne), have been an important topic in space weather and ionospheric research. Neural networks (NNs) emerge as a powerful modeling tool for Ne prediction. However, heavy manual adjustments are time consuming to determine the optimal NN structures. In this work, we propose to use neural architecture search (NAS), an automatic machine learning method, to mitigate this problem. NAS aims to find the optimal network structure through the alternate optimization of the hyperparameters and the corresponding network parameters within a pre‐defined hyperparameter search space. A total of 16‐year data from Millstone Hill incoherent scatter radar (ISR) are used for the NN models. One single‐layer NN (SLNN) model and one deep NN (DNN) model are both trained with NAS, namely SLNN‐NAS and DNN‐NAS, for Ne prediction and compared with their manually tuned counterparts (SLNN and DNN) based on previous studies. Our results show that SLNN‐NAS and DNN‐NAS outperformed SLNN and DNN, respectively. These NN predictions of Ne daily variation patterns reveal a 27‐day mid‐latitude topside Ne variation, which cannot be reasonably represented by traditional empirical models developed using monthly averages. DNN‐NAS yields the best prediction accuracy measured by quantitative metrics and rankings of daily pattern prediction, especially with an improvement in mean absolute error more than 10% compared to the SLNN model. The limited improvement of NAS is likely due to the network complexity and the limitation of fully connected NN without the time histories of input parameters.

Fairness, Bias, and Ethics in AI: Exploring the Factors Affecting Student Performance

The use of artificial intelligence (AI) as a data science tool for education has enormous potential for increasing student performance and course outcomes. However, the growing concern about fairness, bias, and ethics in AI systems requires a careful examination of these issues in an educational context. Using AI and predictive modelling tools, this paper explores the aspects influencing student performance and course success. The Open University Learning Analytics Dataset (OULAD) is analysed using several AI techniques (logistic regression and random forest) in this study to reveal insights about fairness, ethics, and potential biases. This dataset has been used by hundreds of studies to explore how educational data mining can provide information on students. However, potential bias or unfairness in that dataset could undermine the results and any conclusions made from them. To gain insights into the dataset's properties, this was analysed using a typical data science methodology, which included data collecting, cleaning, and exploratory data analysis using Python. By applying AI-based predictive models, this study aims to detect potential biases and their impact on student outcomes. Fairness and ethical considerations are central to the analysis as the representation of various demographic groups and any disparities are evaluated in course results. The goal is to provide useful insights on the proper use of AI in education, while also maintaining equitable and transparent decision-making procedures. The findings shed light on the complicated interplay between artificial intelligence, fairness, and ethics in the context of student performance and course success. As artificial intelligence continues to influence the educational landscape, this study will provide useful ideas for encouraging fairness and minimising biases, resulting in a more inclusive and equal learning environment.

Review on Different Methods for Smart Dam Operation and Water Monitoring

Abstract: This complete literature review digs into the complex world of automated dam operations and water quality management, emphasising key advances and developing techniques in this crucial subject. The study focuses on the integration of sensor technologies, IoT systems, and data analytics to explain how these advances have transformed dam management and water quality preservation. The study gives essential insights into the cutting-edge approaches used in automated dam operations and water quality management by thoroughly examining numerous studies and research findings. Discussions on the use of real-time monitoring systems to track critical water parameters, as well as the use of predictive modelling tools to detect possible problems with water quality early on, are especially relevant. This literature review not only emphasises the importance of technological advancements in improving operational efficiency and environmental stewardship, but it also lays the groundwork for future exploration and refinement of thee methodologies in the pursuit of sustainable water resource management.

Model-driven decision support: A community-based meta-implementation strategy to predict population impact

PurposeStandard tools for public health decision making such as data dashboards, trial repositories, and intervention briefs may be necessary but insufficient for guiding community leaders in optimizing local public health strategy. Predictive modeling decision support tools may be the missing link that allows community level decision makers to confidently direct funding and other resources to interventions and implementation strategies that will improve upon the status quo. MethodsWe describe a community-based model-driven decision support (MDDS) approach that requires community engagement, local data, and predictive modeling tools (agent-based modeling in our case studies) to improve decision-making on implementing strategies to address complex public health problems such as overdose deaths. We refer to our approach as a meta-implementation strategy as it provides guidance to a community on what intervention combinations and their required implementation strategies are needed to achieve desired outcomes. We use standard implementation measures including the Stages of Implementation Completion to assess adoption of this meta-implementation approach. ResultsUsing two case studies, we illustrate how MDDS can be used to support decision making related to HIV prevention and reductions in overdose deaths at the city and county level. Even when community acceptance seems high, data acquisition and diffuse responsibility for implementing specific strategies recommended by modeling are barriers to adoption. ConclusionsMDDS has the capacity to improve community decision makers use of scientific knowledge by providing projections of the impact of intervention strategies under various scenarios. Further research is necessary to assess its effectiveness and the best strategies to implement it.

Predicting the hydraulic response of critical transport infrastructures during extreme flood events

Understanding the effects of extreme floods on critical infrastructures such as bridges is paramount for ensuring safety and resilient design in the face of climate change and extreme events. This study develops robust computational and predictive modeling tools for assessing the impacts of extreme floods on the hydraulic response and structural resilience of bridges. A computational fluid dynamic (CFD) model utilizing RANS equations and k-ω Shear Stress Transport (SST) for simulating supercritical flows is adopted to compute hydrodynamic pressures and the water levels on bridge piers of cylindrical and rectangular shapes during a flood event. The CFD model is validated based on the case study data obtained from the Haj Omran Bridge, built on the Khorramabad River in Iran. The numerical simulations consider hydrological conditions and exclude geotechnical parameters and abutment damages. The numerical results are evaluated based on well-established design guidelines. Machine learning techniques, including Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Regression (SVR), optimized with Grid Search Cross-Validation (GSCV), are adopted to enhance the accuracy of hydrodynamic pressure forecasting at bridge piers. The XGBoost model exhibits superior performance (R2 = 0.908, RMSE = 0.0279, and E = 3.41%) compared to the RF and SVR models. All the estimated pressure data by XGBoost falls within ±6 percent error lines, highlighting the model's robustness for out-of-range hydrodynamic pressure prediction. Additionally, an optimized Long Short-Term Memory (LSTM) model is adopted to effectively predict free surface flow profiles (i.e. flood depth) over the bridge (R2 = 0.937 and RMSE = 0.083), demonstrating its potential for practical applications of flood depth predictions over bridge infrastructures. The proposed methodological framework outlined in this study can facilitate modeling the impacts of extreme floods on bridges, enabling robust climate resilience assessment of critical infrastructures.

Spatio-temporal classification of land use and land cover and its changes in Kerala using remote sensing and machine learning approach.

Land use and land cover (LULC) analysis gives important information on how the region has evolved over time. Kerala, a land with an extensive and dynamic history of land-use changes, has, until now, lacked comprehensive investigations into this history. So the current study focuses on Kerala, one of the ecologically diverse states in India with complex topography, through Landsat images taken from 1990 to 2020 using two different machine learning classifications, random forest (RF) and classification and regression trees (CART) on Google Earth Engine (GEE) platform. RF and CART are versatile machine learning algorithms frequently employed for classification and regression, offering effective tools for predictive modelling across diverse domains due to their flexibility and data-handling capabilities. Normalised Difference Vegetation Index (NDVI), Normalised Differences Built-up Index (NDBI), Modified Normalised Difference Water Index (MNDWI), and Bare soil index (BSI) are integral indices utilised to enhance the precision of land use and land cover classification in satellite imagery, playing a crucial role by providing valuable insights into specific landscape attributes that may be challenging to identify using individual spectral bands alone. The results showed that the performance of RF is better than that of CART in all the years. Thus, RF algorithm outputs are used to infer the change in the LULC for three decades. The changes in the NDVI values point out the loss of vegetation for the urban area expansion during the study period. The increasing value of NDBI and BSI in the state indicates growth in high-density built-up areas and barren land. The slight reduction in the value of MNDWI indicates the shrinking water bodies in the state. The results of LULC showed the urban expansion (158.2%) and loss of agricultural area (15.52%) in the region during the study period. It was noted the area of the barren class, as well as the water class, decreased steadily from 1990 to 2020. The results of the current study will provide insight into the land-use planners, government, and non-governmental organizations (NGOs) for the necessary sustainable land-use practices.

The economics of deep and machine learning-based algorithms for COVID-19 prediction, detection, and diagnosis shaping the organizational management of hospitals

Research background: Deep and machine learning-based algorithms can assist in COVID-19 image-based medical diagnosis and symptom tracing, optimize intensive care unit admission, and use clinical data to determine patient prioritization and mortality risk, being pivotal in qualitative care provision, reducing medical errors, and increasing patient survival rates, thus diminishing the massive healthcare system burden in relation to severe COVID-19 inpatient stay duration, while increasing operational costs throughout the organizational management of hospitals. Data-driven financial and scenario-based contingency planning, predictive modelling tools, and risk pooling mechanisms should be deployed for additional medical equipment and unforeseen healthcare demand expenses. Purpose of the article: We show that deep and machine learning-based and clinical decision making systems can optimize patient survival likelihood and treatment outcomes with regard to susceptible, infected, and recovered individuals, performing accurate analyses by data modeling based on vital and clinical signs, surveillance data, and infection-related biomarkers, and furthering hospital facility optimization in terms of intensive care unit bed allocation. Methods: The review software systems employed for article screening and quality evaluation were: AMSTAR, AXIS, DistillerSR, Eppi-Reviewer, MMAT, PICO Portal, Rayyan, ROBIS, and SRDR. Findings & value added: Deep and machine learning-based clinical decision support tools can forecast COVID-19 spread, confirmed cases, and infection and mortality rates for data-driven appropriate treatment and resource allocations in effective therapeutic and diagnosis protocol development, by determining suitable measures and regulations and by using symptoms and comorbidities, vital signs, clinical and laboratory data and medical records across intensive care units, impacting the healthcare financing infrastructure. As a result of heightened use of personal protective equipment, hospital pharmacy and medication, outpatient treatment, and medical supplies, revenue loss and financial vulnerability occur, also due to expenses related to hiring additional staff and to critical resource expenditures. Hospital costs for COVID-19 medical care, screening, treatment capacity expansion, and personal protective equipment can lead to further financial losses while affecting COVID-19 frontline hospital workers and patients.

Predictive modeling and web-based tool for cervical cancer risk assessment: A comparative study of machine learning models

In today's digital era, the rapid growth of databases presents significant challenges in data management. In order to address this, we have developed and designed CHAMP (Cervical Health Assessment using machine learning for Prediction), which is a user interface tool that can effectively and efficiently handle cervical cancer databases to detect patterns for future prediction diagnosis. CHAMP employs various machine learning algorithms which include XGBoost, SVM, Naive Bayes, AdaBoost, Decision Tree, and K-Nearest Neighbors in order to predict cervical cancer accurately. Moreover, this tool also designates to evaluate and optimize processes, to retrieve the significantly augmented algorithm for predicting cervical cancer. Although, the developed user interface tool was implemented in Python 3.9.0 using Flask, which provides a personalized and intuitive platform for pattern detection. The current study approach contributes to the accurate prediction and early detection of cervical cancer by leveraging the power of machine learning algorithms and comprehensive validation tools, which aim to provide learned decision-making.•CHAMP is a user interface tool which is designed for the detection of patterns for future diagnosis and prognosis of cervical cancer.•Various machine learning algorithms are employed for accurate prediction.•This tool provides personalized and intuitive data analysis which enables informed decision-making in healthcare.

Credit risk prediction based on causal machine learning: Bayesian network learning, default inference, and interpretation

AbstractThe predictive and interpretable power of models is crucial for financial risk management. The purpose of this study was to perform credit risk prediction in a structured causal network with four stages—data processing, structural learning, parameter learning, and interpretation of inferences—and use six real credit datasets to conduct empirical research on the proposed model. Compared with traditional machine learning algorithms, we comprehensively explain the results of credit default through forward and reverse reasoning. We also compared our model with the post hoc interpretation models local interpretable model‐agnostic explanations (LIME) and shapley additive explanations (SHAP) to verify the interpretability of Bayesian networks. The experimental results show that the prediction performance of Bayesian networks is superior to traditional machine learning models and similar to the performance of ensemble models. Furthermore, Bayesian networks offer valuable insights into the interplay of features by considering their causal relationships and enable an assessment of how individual features influence the prediction outcome. In this study, what‐if analysis was performed to assess credit default probabilities under various conditions. This analysis provides decision‐makers with the necessary tools to make informed judgments about the risk profile of borrowers. Consequently, we consider Bayesian networks as a viable tool for credit risk prediction models in terms of prediction performance and interpretability.

Finite element modeling in heat and mass transfer of potato slice dehydration, nonisotropic shrinkage kinetics using arbitrary Lagrangian–Eulerian algorithm and artificial neural network

AbstractThe present study optimized the drying temperature (50–80°C) for potato slices based on color, texture, and visual observations. At an optimized temperature (60°C), a 2D axisymmetric finite element method (FEM) was developed in COMSOL Multiphysics to predict the heat and mass transfer (HMT) in a disk‐shaped potato slice. The nonisotropic shrinkage was predicted for the potato slice by the arbitrary Lagrangian–Eulerian approach. The experimental dehydration results revealed that axial shrinkage (27.44%) was 2.5 times higher than radial shrinkage (67.39%). The simulated outcomes based on FEM revealed the realistic visualization of spatial heat transfer, moisture migration, and nonisotropic slice deformation. The predicted moisture content, surface temperature, and shrinkage properties were in good agreement with the experimental results. The shrinkage behavior was further validated using artificial neural network (ANN) to simulate the slice shrinkage. Results showed that both the COMSOL and ANN approaches can precisely predict the shrinkage‐dependent HMT model. The ANN model outperformed the COMSOL determined by mean absolute error, mean square error (MSE), root MSE, and Chi‐square (χ2) values. The successful application of the presented approach for determining dehydration characteristics may have potential for quality assessment and management of different fruits and vegetables.Practical applicationsThis journal article explores the practical industrial applications of combining finite element method (FEM)‐based heat and mass transfer model and artificial neural networks (ANNs) to improve the efficiency and quality of food drying processes. FEM is employed to simulate and predict the realistic visualization of heat and mass transfer phenomena along with non‐isotropic shrinkage, while ANN serves as a data‐driven modeling tool for process control and prediction. The integration of these two technologies offers significant advantages in the food industry, including quantification of precise temperature and moisture content, as well as to monitor the drying process of various food products, reduced energy consumption, and time.

Development of A Neural Boosted Model and JSL Code to Identify “Clean” or “Not Clean” Wells - A West Texas Sperry and Oklahoma Woodford Fractured Wells Coiled Tubing Cleaning Case Study

In a previous study, wellbore cleaning coefficient (WCC) correlations for cleaned wellbores out of debris and bridge plug remnants were developed for three conventional coiled tubing sizes (2.375”, 2.625”, and 2.875”). The following key performance indicators (KPIs): (1) slick water density ( ) ρ f , (2) slick water viscosity ( ) µ f , (3) hydraulic diameter c t (d - d ) between casing inner diameter (dc ) and coil tubing outer diameter (dt ), (4) average annular velocity ( ) v and (5) cleaning pressure gradient ∆P across a measured depth (MD) were employed in the empirical models. The models addressed operational conditions under which fractured wells will be identified as whether “clean” or “not clean”. In this study, the database from 150 wells, in the Spraberry formation in West Texas, was used to develop a predictive model to identify status of cleaned fractured wells: whether “clean” or “not clean”? About 70% of the data (99 wells) was used for training and about 30% (51 wells) for validation. 14 wells from the liquids-rich shale Woodford formation (Oklahoma) were utilized for testing. Six predictive modeling tools were designed to validate the derived empirical correlations. These tools are (1) Fit Stepwise, (2) Neural Boosted, (3) Boosted Tree, (4) Decision Tree (Partition), (5) Generalized Regression Lasso, and K-Nearest Neighbors. In the predictive models, independent variables are the annular velocity (AV), the Reynolds’ Number (Re), the Euler’s Number (Eu), and the coiled tubing roughness to internal radius ratio (ε/D). The dependent variable is well status; “clean” or “not clean”. Jump Scripting Language (JSL) code was used to develop user-friendly software. The software would be utilized to identify the fractured wellbore status, whether “clean” or “not clean”. Operators would be able to use the code to identify working conditions for which completed fractured wells are “clean” out of fracturing debris and remnants of bridge plugs or “not clean”. Input parameters to the code are AV, Re, Eu, and ε/D

Prediction and control of spatiotemporal chaos by learning conjugate tubular neighborhoods

I present a data-driven predictive modeling tool that is applicable to high-dimensional chaotic systems with unstable periodic orbits. The basic idea is using deep neural networks to learn coordinate transformations between the trajectories in the periodic orbits’ neighborhoods and those of low-dimensional linear systems in a latent space. I argue that the resulting models are partially interpretable since their latent-space dynamics is fully understood. To illustrate the method, I apply it to the numerical solutions of the Kuramoto–Sivashinsky partial differential equation in one dimension. Besides the forward-time predictions, I also show that these models can be leveraged for control.

Instrumental and methodological support for the assessment and forecasting of knowledge in the pedagogical process

Currently, ensuring the process of assessing students' knowledge is associated with the development of adaptive instrumental and methodological support, which allows you to take into account and adjust the level of training of each student. Machine learning methods and artificial neural networks are widely used in solving problems of modeling and instrumental support of pedagogical processes. It should be noted that effective results can be obtained using multinomial logistic regression based on an integrated approach combining interactive testing and predictive modeling tools. The purpose of the article is to develop instrumental and methodological support for the assessment and forecasting of knowledge in the pedagogical process, taking into account the use of interactive testing algorithms and predictive modeling based on multinomial logistic regression. A software package has been developed to assess students' knowledge and predict academic performance in the educational process, which combines interactive testing tools with predictive modeling using multinomial logistic regression. For the development of the software package, the high-level Python language was used with the involvement of mathematical libraries. The aspects of the development of a support system for the course "Intelligent transport systems" in higher and additional education systems are considered. The obtained results can be used in solving problems of improving intellectual educational environments and in modeling pedagogical processes. In addition, the results can be used to analyze the effectiveness of teaching methods, identify trends and patterns in the educational process, as well as predict student academic performance.

Energy Efficiency in Buildings using Internet of Things (IoT) and Big Data Analytics

In order to assess the energy efficiency of building activities in real-time, this research offers a data-driven methodology. Efficiently managing building energy usage while minimizing negative effects on the environment is the focus of this study. Using a large dataset that includes efficiency ratings obtained from sophisticated analytics and continuous monitoring, as well as specific energy consumption (SEC) measurements, our study reveals intricate patterns in energy use. Reducing energy consumption by 15% during peak hours is possible with the use of predictive modeling tools, which show the possibility of proactive actions. With dynamic modifications resulting in a 20% reduction in total energy use, there are substantial benefits to implementing adaptive techniques based on real-time data. The method’s dependability is confirmed by comparing it to industry-standard standards, which highlights how strong the real-time evaluation system is. Building managers may benefit greatly from this research’s findings on energy efficiency, which will help to create more sustainable and financially feasible building systems.

Challenges in predicting and managing construction noise impacts in urban environments. Case studies from Sydney, Australia

Approval of major infrastructure projects in Australia is contingent on a vigorous environmental assessment process demonstrating that the impacts from such proposals are acceptable and can be satisfactorily managed. As part of this process, an assessment of construction noise is undertaken by predicting the level and duration of impact. Given the scale, complexity and programs for delivery of major infrastructure projects such as the Sydney Metro, the approach to assessment of environmental impacts is typically undertaken based on early concept designs, and assumptions which enable potential worst-case impacts to be considered. Ongoing design development and construction methodologies continue to occur, and are finalised by the contractor responsible for delivering the package of works for the project. Layers of conservatism could therefore be built into the process to ensure that environmental impacts are understood, and appropriate mitigation measures are considered. Moreover, common predictive noise modelling tools are believed to under-estimate the level of barrier insertion loss provided by residential buildings in urban scenarios, further exacerbating the over-prediction of construction noise impacts. By use of case studies, the present paper examines the challenges of assessing and managing construction noise associated with a major infrastructure project in Sydney, Australia.

Reverse Engineering of Antigenic Peptides in LGL to Decipher Disease Triggers

One of the hypotheses as to the pathogenesis of T-cell large granular lymphocytic leukemia (LGL) posits its origin from the initially polyclonal stimulation by unknown antigens, e.g., viral, autoimmune or tumor antigens. LGL evolves then via oligoclonal and ultimately monoclonal cytotoxic T-cell expansion. This process may be accompanied by acquisition of STAT3 (or rarely STAT5) mutations in 40% of cases. Identification of driver antigens is a key research question in LGL as it could both clarify its triggers as well as the mechanisms of cytopenias. We investigated a cohort of patients with this disease (n=195) by combining HLA allotypes and NGS-based TCR analysis to identify marker TCR Vβ clonotypes with the goal to reverse engineer or deduce peptides and therefore antigenic targets. For this purpose we used in silico and machine-learning modelling to capture structural features predictive of CDR3 Vβ epitope interactions in specific HLA allotype context. To focus on specific HLA/TCR interactions we have analyzed HLA for the presence of common haplotypes. First, we generated HLA haplotype calls using whole-exome sequencing data. We have found that HLA-A*02 was the most common class I allele, but not overrepresented in LGL vs controls. We have also identified HLA-B*44 as a risk allele present in 36% LGL with STAT3 mutation which was further enriched when considering HLA-A*02/HLA-B*44 together (Fig.1A). After TCR NGS we selected a threshold of 5.0% (based on study of 900 controls) to define an expanded (co)dominant clonotypes: 145 expanded clones were identified, of which 143 were unique/private. To identify possible consensus sequence across the CDR3 profile, we clustered the CDR3 AA sequences based on global alignment scores (Needleman-Wunsch) using BLOSUM62 substitution matrix and identified 9 clusters. However, multiple-sequence alignment within each CDR3 cluster did not show any distinct consensus sequence patterns overall and within cases sharing HLA-A*02. Cross-referencing with CDR3 databases of known conditions yielded 87 CDR3β (co)dominant LGL clonotypes. About 50% of those LGL clonotypes were shared in 2 common disease: type I diabetes, Celiac disease We also mined the VDJdb[2] for known CDR3-pHLA associations. Filtering to include high quality annotations, matches and HLA-A*02 restriction; LGL-derived TCR were found to recognize multiple epitopes. Prominent examples include CMV and EBV derived peptidic antigens from pp65, BRLF1, and BMLF1 (NLVPMVATV, RPPIFIRRL) present in 44% LGL expanded clonotypes (Fig.1B). In order to reverse engineer the identity of antigenic peptides based on the CDR3β sequence and HLA type (high affinity peptides restricted to HLA A2, a machine-learning method was developed as predictive modelling tool that quantifies CDR3β/peptide binding. The previously developed LLM trained on AA sequences was used for that purpose[3] whereby LLM embeddings serve as input to a deep CNN-based model, which is trained on IEDB pos/neg datasets[4]. This method was augmented with control CDR3 sequences as negative data points. As a result, our model showed an AUC of 0.94 in cross-validation. As a proof of concept of the proposed model in epitope discovery, CMV pp65-derived peptides were fitted into specific LGL CDR3 and CMV specific CDR3 clonotypes were found to match e.g., pp65 epitope NLVPMVATV among several others overlapping with the above-described findings. In conclusion, our analysis shows that a significant number of expanded clonotypes may be derived from known epitopes e.g., in autoimmune disease and viral infection pointing towards molecular mimicry. However, the specificity of a large fraction of the LGL spectrum remains unknown. For the discovery of corresponding peptides we generated ML-based antigen simulation method allowing to identify the best fitting peptides into the corresponding HLA allele and CDRVβ clonotype. Fig. Global map of co-occurrence of HLA allotypes showing the delta frequency compared to healthy controls and Sankey plot of identified CDR3β - Epitope associations (a) Using the whole-exome sequencing based HLA genotype calls, we calculated the frequency of co-occurrence of different HLA loci and investigated the differences in regards to controls (b) Using the VDJdb, we conducted sequence similarity based search and extracted the CDR3-pHLA annotations.