BIFI-TALK: Xavier de la Cruz Monserrat. Vall d’Hebron Institute of Research, Barcelona

Title: Cracking the Hereditary Disease Code: Revisiting Predictive Properties with a Probabilistic Touch

Abstract

As genomic sequencing becomes widespread in clinical settings, in silico pathogenicity predictions are increasingly vital for precision medicine. However, despite advances in machine learning-based classifiers, existing approaches are reaching a performance plateau while still facing interpretation challenges. In response, a paradigm shift has emerged in recent years, moving from binary classification toward modeling the quantitative impact of genetic variants.
Our work in this area has shown good performance, as demonstrated by our contributions to the CAGI5 and CAGI6 challenges, yet there is still room for improvement. In this talk, I will focus on a key aspect of pathogenicity prediction: the extraction of new discriminant features from traditional predictive properties, using a simple probabilistic framework. In particular, I will share insights from our approach and present preliminary results that highlight its potential impact on predictive models.

Friday, 14 February 2025, 12:30 h
Conference Room. Edificio I+D