Phaseek: A Generalizable Predictor of Liquid–Liquid Phase Separation from Protein Sequence

This dataset provides predicted liquid–liquid phase separation (LLPS) propensity scores across the full proteomes of 18 widely studied model organisms.

The predictions were generated using Phaseek, our deep learning model developed to identify sequence features associated with LLPS behavior.

For each protein, we include residue-level LLPS scores and highlight key regions predicted to drive phase separation, alongside randomly selected regions for comparison. The dataset also includes physicochemical properties and secondary structure features for both key and random regions, as described in the Phaseek publication. These LLPS-driving segments may correspond to functional elements involved in biomolecular condensate formation.

The organisms and their corresponding abbreviations are:

• Hs – Homo sapiens

• Mm – Mus musculus

• Rn – Rattus norvegicus

• Dm – Drosophila melanogaster

• Dr – Danio rerio

• At – Arabidopsis thaliana

• Sc – Saccharomyces cerevisiae

• Ce – Caenorhabditis elegans

• Bt – Bos taurus

• Mmu – Macaca mulatta

• Cf – Canis lupus familiaris

• Ss – Sus scrofa

• Gg – Gallus gallus

• Xl – Xenopus laevis

• EcK12 – Escherichia coli K-12

• Pt – Pan troglodytes

• Ag – Anopheles gambiae

• Pf – Plasmodium falciparum

This dataset was generated using the Phaseek predictor and supports the findings presented in our publication:
Generalizable Prediction of Liquid–Liquid Phase Separation from Protein Sequence
http://dx.doi.org/10.1101/2025.01.27.635039