Predicting protein condensate formation using machine learning

Guido van Mierlo, Jurriaan Jansen, Simon van Heeringen and Michiel Vermeulen, theme Cancer development, and colleagues recently published in Cell Reports that predicting protein condensate formation can be done using machine learning.

Membraneless organelles are liquid condensates, which form through liquid-liquid phase separation. Recent advances show that phase separation is essential for cellular homeostasis by regulating basic cellular processes, including transcription and signal transduction. The reported number of proteins with the capacity to mediate protein phase separation (PPS) is continuously growing.

While computational tools for predicting PPS have been developed, obtaining a proteome-wide overview of PPS probabilities has remained challenging. Here, they present a phase separation analysis and prediction (PSAP) machine-learning classifier that, based solely on the amino acid content of a training set of known PPS proteins, can determine the phase separation likelihood for each protein in a given proteome. 

Through comparison with PPS databases, existing predictors, and experimental evidence, they demonstrate the validity and advantages of the PSAP classifier. They anticipate that the PSAP predictor provides a useful tool for future research aimed at identifying phase separating proteins in health and disease.