This task will oversee all processes related to data management throughout the project. These processes include data input, data modelling and standardization, secure data storage, anonymous medical data handling, and High Performance Computing (HPC). Initially, anonymized data for which there are signed consent forms will be stored in a secure storage in accordance with GDPR and national regulations.

This task will transform and homogenize the data formats from the different sources and systems of the Neutro-NARPS repositories into a structured format for downstream analyses. For this purpose, appropriate models [44, 45] suitable for the application of a set of common computational tasks in dispersed or incomplete databases commonly seen in medical questionnaires will be used. During the data pre-processing the non-ubiquitous patient attributes will be eliminated increasing the cohort’s study strength.

This task will oversee the application of state-of-the-art statistical analyses and clarification of the interplay of factors that govern the potential development of CH, CVD, AID and cancer in CNP patients with various clinical, biometric and mutational signatures from WG1, WG2. Differences in group proportions associated with inflammation, CVD, CH, AID and cancer patterns will be evaluated between CNP and healthy individuals. Survival analyses will be performed and differences between groups will be evaluated. Multivariable models will be used to estimate the hazard ratios of CNP for the association of overall survival and different risk variables for CH, CVD, cancer and inflammation.

This task will unlock the full potential of ML, encompassing multi-source data from genomics, demographics and clinical parameters towards CNP risk prediction. The task is divided in two sub-tasks with the first being on feature selection and the second on data preparation for subsequent ML Analytics. Feature selection is based on the application of statistical algorithms to identify significant biomarkers that will be then incorporated into the ML model. The discovered biomarkers will be used as input features along with patient diagnosis information to train a configurable suite of supervised ML classifiers [46,47] for (a) studying the relationships between demographic, nutritional, multi-omics, imaging and immunological variants and phenotypes of CNP patients and (b) defining risk profiles for CNP patients with relative prognosis. Interpretability and explainability of the ML model will be addressed to support fairness (e.g., from statistical parity) and trustworthiness (e.g., mitigate bias and variability from the different centres, etc.).