Status: Ongoing
About the project
The data revolution is changing the foundations of biomedicine radically, so that modern biostatistics is becoming one of the fundamental pillars of research success. The internet, sensors on smartphones and wearable devices, and cheap data transfers make it easy to collect data rapidly and extensively. Cheaply collected data bring valuable information content in various contexts, from patient safety to improving businesses. Yet, biases and incompleteness lead to poor representation of complexity and structure, thus calling for radical new methods. BRIDGE focuses on the development of statistical methods for the integration of multiple data sources arising in widely different contexts. An important focus of the project is on biomedical applications, which demand reliable methods and reproducible results for clinical implementation and patient safety.
The frontier of digital medicine and economy is providing accurate predictions for single individuals (patients or users). To achieve this overall objective, and starting from complex multi-source data, BRIDGE will combine:
1. innovative rank-based methods, as they well represent individual behaviours and preferences, and are more robust;
2. algorithms that scale to ultra-large data dimensions, as this is required for analysis of modern data;
3. real-time machine learning techniques, to achieve fast targeted decision-making.
BRIDGE will thus lay the foundations of the first comprehensive statistical approach to rank data integration, which also quantifies all uncertainties in the models used for prediction as well as in the results. Accounting for uncertainty in estimation gives a crucial advantage, since knowledge of the reliability of suggested actions has a major impact on decision-making. Finally, BRIDGE will also develop far-reaching implementation solutions, tested in the context of both cancer genomics and internet-based marketing.
Primary objectives
The primary objective of BRIDGE is developing a flexible scalable unsupervised rank-based Bayesian framework, suited to the integration of multiple data sources in ultra-high dimensional data situations, to facilitate both signature discovery in molecular medicine, for reliable and reproducible genomics pipelines, and preference learning from sparse and incomplete data, for enhanced recommender systems. This overarching objective entails the following secondary objectives:
O1 Scalability: Development & validation of novel algorithmic tools for ultra-high dimensional data.
O2 Feature-informed: Integration of exogenous sources of information with diverse characteristics, and application to the enhancing of the fine-tuning of LLMs (sub-aim of O2).
O3 Integration of heterogeneous data sources. Joint inferential solution to handle several correlated data layers with varying structure.
O4 On-line learning. Modelling time-dynamics, and specifying it to perform on-line inference and prediction.
Project team
- PI Associate Professor Valeria Vitelli, UiO, Institute of Basic Medical Sciences
- Associate Professor Anne-Marie George, UiO, Department of Informatics
Funding
Research Council of Norway, FRIPRO programme for Researcher Project for Experienced Scientists
Project number: 355446
Project period: 1 June 2025 - 31 May 2029