Abstract:

High-throughput technologies are becoming increasingly important in
discovering prognostic biomarkers and in identifying novel drug targets.
With various prognostic molecular signatures, cancer serves as a
paradigmatic example of the utility of high-throughput data in
identifying prognostic biomarkers, often distilled into relatively short
gene lists. Such gene lists can be obtained as a set of features (genes)
that are important for the decisions of a Machine Learning (ML) method
applied to high-dimensional gene expression data. This high-dimensional
data can be structured by graphs, which are very common in the
biomedical domain. For instance, a patient can be represented by a
protein-protein interaction (PPI) network where the nodes contain the
patient-specific omics features. Graph Neural Networks (GNNs) can
classify these graph-based representations.

Explanation methods applied to GNNs produce explanations of individual
predictions that can be utilized to construct patient-specific
subnetworks. When aggregated across patients, these explanations enable
model-wide feature selection. In this talk, I will discuss both aspects
and present a methodology to: (i) derive patient-specific subnetworks
that are potentially valuable for precision medicine approaches, and
(ii) systematically and quantitatively analyze the stability, impact on
classification performance, and biological interpretability of the
model-wide selected feature sets. Finally, I will present our
Ensemble-GNN approach for classifying graph-structured data, which can
be used to deploy federated, ensemble-based GNNs in Python.

Tüm Etkinlikler