Beta turn propensity and a model polymer scaling exponent identify intrinsically disordered phase-separating proteins

Abstract

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Intrinsically disordered regions that drive phase separation form a robustly distinct protein class

“…A limitation of the previous work, including our own (42), has been the relatively small set of sequences used to train predictors. We first sought to alleviate this problem by identifying additional sequences in our two negative control categories, folded proteins and IDRs, which are not thought to phase separate.…”

“…We previously developed a predictive model of LLPS behavior, ParSe (“ Par tition Se quence”), that identifies phase-separating (PS) IDRs starting from predictions of hydrodynamic size, which is indicative of the relative strength of intramolecular as compared to solvent interactions (42). The ParSe algorithm uses a sequence-calculated polymer scaling exponent, v model , to quantify hydrodynamic size (43, 44).…”

“…The ParSe algorithm uses a sequence-calculated polymer scaling exponent, v model , to quantify hydrodynamic size (43, 44). When v model is combined with a second sequence-based parameter, the intrinsic propensity for a sequence to form β-turns (45), the algorithm can distinguish between sequences belonging to one of three classes of protein regions: folded, ID, and PS ID (42). We proposed a physical mechanism whereby transient β-turn structures reduce the desolvation penalty of forming a protein-rich phase and increase exposure of atoms involved in π/sp 2 valence electron interactions.…”

“…We proposed a physical mechanism whereby transient β-turn structures reduce the desolvation penalty of forming a protein-rich phase and increase exposure of atoms involved in π/sp 2 valence electron interactions. In this mechanism, β-turns could act as energetically favored nucleation points, potentially explaining the observed higher propensity for turns in IDRs that drive phase separation in vivo (42, 45).…”

An interpretable machine learning algorithm to predict disordered protein phase separation based on biophysical interactions

“…Another, ParSe, combines two physical features, hydrodynamic size of monomeric proteins and beta-turn propensity estimated from polymer models, to predict phase separation propensity, however uses only composition and not residue context when making predictions. [43] A third, PSAP, uses compositional bias of phase-separating proteins and 10 amino-acid biochemical features as features for a random forest classifier with a 0.89 AUROC (area under the receiver operating characteristics curve), yet also lacks residue context in the prediction. [44] A major issue in developing a phase separation predictor is the selection of a negative training set.…”