Linear Interacting Peptides (LIPs) detection in protein sequences based on embedding models

This work explores the ability of embedding models of representing essential structural and biological features of proteins, specifically focusing on Linear Interacting Peptides (LIPs), a class of binding disordered regions recently introduced in the MobiDB database. The aim of this work was to test a LIPs detection and classification method using learning models and starting from protein sequences. The first part of this work is centered on the creation of a comprehensive target dataset for training the model, built upon the outputs from FLIPPER. This data was then further filtered by combining both disorder and binding information. A second part is dedicated to the training of a CNN model for LIPs discrimination.