Project start: 1.10.2025.

Liquid–liquid phase separation (LLPS) is a key mechanism responsible for the formation of dynamic droplet-like condensates that regulate various cellular functions and enable the development of novel biomaterials. Although LLPS has been extensively studied at the protein level, predicting and understanding LLPS in peptides represents a novel and challenging research direction. As shorter and structurally simpler biomolecules, peptides offer unique opportunities for designing functional materials based on phase separation principles. However, exploring their chemical space poses a major combinatorial challenge due to the vast number of possible sequences, and the complex and still insufficiently understood rules governing this process limit the applicability of standard LLPS prediction approaches.

The project aims to develop a computational framework based on machine learning and evolutionary computing for the discovery of peptides with a high propensity for LLPS. By developing deep learning models, the project will explore knowledge transfer from proteins to peptides to improve the classification and interpretation of LLPS propensity. Emphasis will be placed on explainability techniques to uncover patterns within peptide sequences and integrate this knowledge into multi-objective generative models capable of proposing novel functional peptides. Experimental validation of generated candidates will further confirm the effectiveness of the developed approach.

This project explores the possibility of mining knowledge from peptide and protein datasets, opening potential applications in other functional areas such as therapeutic or catalytic activity. To realize this ambitious goal, a multidisciplinary team of experts in computer science, AI, peptide chemistry and biochemistry has been assembled. The project has the potential to advance our understanding of LLPS in peptides and engineering of new nanomaterials with potential applications in biotechnology and medicine.