ADOPT: intrinsic protein disorder prediction through deep bidirectional transformers.

Intrinsically disordered proteins (IDPs) are important for a broad range of biological functions and are involved in many diseases. An understanding of intrinsic disorder is key to develop compounds that target IDPs. Experimental characterization of IDPs is hindered by the very fact that they are highly dynamic. Computational methods that predict disorder from the amino acid sequence have been proposed. Here, we present ADOPT (Attention DisOrder PredicTor), a new predictor of protein disorder. ADOPT is composed of a self-supervised encoder and a supervised disorder predictor. The former is based on a deep bidirectional transformer, which extracts dense residue-level representations from Facebook's Evolutionary Scale Modeling library. The latter uses a database of nuclear magnetic resonance chemical shifts, constructed to ensure balanced amounts of disordered and ordered residues, as a training and a test dataset for protein disorder. ADOPT predicts whether a protein or a specific region is disordered with better performance than the best existing predictors and faster than most other proposed methods (a few seconds per sequence). We identify the features that are relevant for the prediction performance and show that good performance can already be gained with <100 features. ADOPT is available as a stand-alone package at https://github.com/PeptoneLtd/ADOPT and as a web server at https://adopt.peptone.io/.

pepKalc: scalable and comprehensive calculation of electrostatic interactions in random coil polypeptides.

Motivation: Polypeptide sequence length is the single dominant factor hampering the effectiveness of currently available software tools for de novo calculation of amino acid-specific protonation constants in disordered polypeptides. Results: We have developed pepKalc, a robust simulation software for the comprehensive evaluation of protein electrostatics in unfolded states. Our software completely removes the limitations of the previously reported Monte-Carlo approaches in the computation of protein electrostatics by using a hybrid approach that effectively combines exact and mean-field calculations to rapidly obtain accurate results. Paired with a modern architecture GPU, pepKalc is capable of evaluating protonation behavior for an arbitrary-size polypeptide in a sub-second time regime. Availability and implementation: http://protein-nmr.org and https://github.com/PeptoneInc/pepkalc.

Using NMR chemical shifts to calculate the propensity for structural order and disorder in proteins.

NMR spectroscopy offers the unique possibility to relate the structural propensities of disordered proteins and loop segments of folded peptides to biological function and aggregation behaviour. Backbone chemical shifts are ideally suited for this task, provided that appropriate reference data are available and idiosyncratic sensitivity of backbone chemical shifts to structural information is treated in a sensible manner. In the present paper, we describe methods to detect structural protein changes from chemical shifts, and present an online tool [ncSPC (neighbour-corrected Structural Propensity Calculator)], which unites aspects of several current approaches. Examples of structural propensity calculations are given for two well-characterized systems, namely the binding of α-synuclein to micelles and light activation of photoactive yellow protein. These examples spotlight the great power of NMR chemical shift analysis for the quantitative assessment of protein disorder at the atomic level, and further our understanding of biologically important problems.

ncIDP-assign: a SPARKY extension for the effective NMR assignment of intrinsically disordered proteins.

UNLABELLED: We describe here the ncIDP-assign extension for the popular NMR assignment program SPARKY, which aids in the sequence-specific resonance assignment of intrinsically disordered proteins (IDPs). The assignment plugin greatly facilitates the effective matching of a set of connected resonances to the correct position in the sequence by making use of IDP random coil chemical shifts. AVAILABILITY: The ncIDP-assign extension is available at http://www.protein-nmr.org/.

Sequence-specific random coil chemical shifts of intrinsically disordered proteins.

Although intrinsically disordered proteins (IDPs) are widespread in nature and play diverse and important roles in biology, they have to date been little characterized structurally. Auspiciously, intensified efforts using NMR spectroscopy have started to uncover the breadth of their conformational landscape. In particular, polypeptide backbone chemical shifts are emerging as powerful descriptors of local dynamic deviations from the "random coil" state toward canonical types of secondary structure. These digressions, in turn, can be connected to functional or dysfunctional protein states, for example, in adaptive molecular recognition and protein aggregation. Here we describe a first inventory of IDP backbone (15)N, (1)H(N), (1)H(α), (13)C(O), (13)C(ß), and (13)C(α) chemical shifts using data obtained for a set of 14 proteins of unrelated sequence and function. Singular value decomposition was used to parametrize this database of 6903 measured shifts collectively in terms of 20 amino acid-specific random coil chemical shifts and 40 sequence-dependent left- and right-neighbor correction factors, affording the ncIDP library. For natively unfolded proteins, random coil backbone chemical shifts computed from the primary sequence displayed root-mean-square deviations of 0.65, 0.14, 0.12, 0.50, 0.36, and 0.41 ppm from the experimentally measured values for the (15)N, (1)H(N), (1)H(α), (13)C(O), (13)C(ß), and (13)C(α) chemical shifts, respectively. The ncIDP prediction accuracy is significantly higher than that obtained with libraries for small peptides or "coil" regions of folded proteins.

Probing the free energy landscape of the FBP28WW domain using multiple techniques.

The free-energy landscape of a small protein, the FBP 28 WW domain, has been explored using molecular dynamics (MD) simulations with alternative descriptions of the molecule. The molecular models used range from coarse-grained to all-atom with either an implicit or explicit treatment of the solvent. Sampling of conformation space was performed using both conventional and temperature-replica exchange MD simulations. Experimental chemical shifts and NOEs were used to validate the simulations, and experimental phi values both for validation and as restraints. This combination of different approaches has provided insight into the free energy landscape and barriers encountered by the protein during folding and enabled the characterization of native, denatured and transition states which are compatible with the available experimental data. All the molecular models used stabilize well defined native and denatured basins; however, the degree of agreement with the available experimental data varies. While the most detailed, explicit solvent model predicts the data reasonably accurately, it does not fold despite a simulation time 10 times that of the experimental folding time. The less detailed models performed poorly relative to the explicit solvent model: an implicit solvent model stabilizes a ground state which differs from the experimental native state, and a structure-based model underestimates the size of the barrier between the two states. The use of experimental phi values both as restraints, and to extract structures from unfolding simulations, result in conformations which, although not necessarily true transition states, appear to share the geometrical characteristics of transition state structures. In addition to characterizing the native, transition and denatured states of this particular system in this work, the advantages and limitations of using varying levels of representation are discussed.

Deeply branching c6-like cytochromes of cyanobacteria.

The cyanobacterium Synechococcus sp. PCC 7002 carries two genes, petJ1 and petJ2, for proteins related to soluble, cytochrome c6 electron transfer proteins. PetJ1 was purified from the cyanobacterium, and both cytochromes were expressed with heme incorporation in Escherichia coli. The expressed PetJ1 displayed spectral and biochemical properties virtually identical to those of PetJ1 from Synechococcus. PetJ1 is a typical cytochrome c6 but contains an unusual KDGSKSL insertion. PetJ2 isolated from E. coli exhibited absorbance spectra characteristic of cytochromes, although the alpha, beta, and gamma bands were red-shifted relative to those of PetJ1. Moreover, the surface electrostatic properties and redox midpoint potential of PetJ2 (pI 9.7; E(m,7) = 148 +/- 1.7 mV) differed substantially from those of PetJ1 (pI 3.8; E(m,7) = 319 +/- 1.6 mV). These data indicate that the PetJ2 cytochrome could not effectively replace PetJ1 as an electron acceptor for the cytochrome bf complex in photosynthesis. Phylogenetic comparisons against plant, algal, bacterial, and cyanobacterial genomes revealed two novel and widely distributed clusters of previously uncharacterized, cyanobacterial c 6-like cytochromes. PetJ2 belongs to a group that is distinct from both c6 cytochromes and the enigmatic chloroplast c 6A cytochromes. We tentatively designate the PetJ2 group as c6C cytochromes and the other new group as c6B cytochromes. Possible functions of these cytochromes are discussed.