Descripció del projecte
This research project will participate in the development of a new foundational graph model (Graph Foundational Model) that aims to learn on a large scale the biological processes related to cancer from multi-omic biomedical data. This foundational model will allow to apply all this knowledge to different areas of cancer research, without the need for retraining (zero-shot) or with a minimum adjustment (fine-tunning).
Among other factors, this project will address a fundamental problem in biomedical datasets, which is the problem known as HDLSS (High-dimension, low-sample-size). The HDLSS problem appears when the number of samples (n) is much smaller than the number of attributes (d). This causes poor generalization of classical machine learning models due to overfitting and lack of statistical support. This is a very common scenario in the field of biomedicine, where in most cases it is very difficult to collect a large number of samples, but each of them is described with hundreds or even thousands of attributes. This situation makes the traditionally used tabular representation not the most suitable for this environment.
To demonstrate the generality of the proposed solution, the project will also address the application of this same architecture in other environments, such as computer networks and cybersecurity, where similar challenges arise.
