Direct Prediction of Intrinsically Disordered Protein Conformational Properties from Sequence

ABSTRACT

Intrinsically disordered proteins and protein regions (IDPs/IDRs - hereafter IDRs) are ubiquitous across all domains of life. Unlike their folded domain counterparts, IDRs sample a diverse ensemble of conformations - rapidly interconverting between heterogeneous states. Analogous to how folded proteins adhere to a sequence-structure-function relationship, IDRs follow a sequence-ensemble-function paradigm. While experimental methods to study the conformational properties of IDRs exist, they can be challenging, time-consuming, and often require specialized equipment and expertise. Recent methodological advances in biophysical modeling offer a unique opportunity to explore sequence ensemble relationships; however, these methods are often limited in throughput and require both software and technical expertise. In this work, we integrated rational sequence design, large-scale molecular simulations, and deep learning to develop ALBATROSS, a deep learning model for predicting IDR ensemble dimensions from sequence. ALBATROSS is lightweight, easy to use, and readily accessible as both a locally-installable software package, as well as a point-and-click style interface in the cloud. We first demonstrate the applicability of our predictors by examining the generalizability of sequence-ensemble relationships in IDRs. Then, we leverage the high-throughput nature of our networks to characterize emergent biophysical behavior of local and global IDR ensemble features across the human proteome.