A3D database: structure-based predictions of protein aggregation for the human proteome

Abstract

Summary: Protein aggregation is associated with many human disorders and constitutes a major bottleneck for producing therapeutic proteins. Our knowledge of the human protein structures repertoire has dramatically increased with the recent development of the AlphaFold (AF) deep-learning method. This structural information can be used to understand better protein aggregation properties and the rational design of protein solubility. This article uses the Aggrescan3D (A3D) tool to compute the structure-based aggregation predictions for the human proteome and make the predictions available in a database form. In the A3D database, we analyze the AF-predicted human protein structures (for over 20.5 thousand unique Uniprot IDs) in terms of their aggregation properties using the A3D tool. Each entry of the A3D database provides a detailed analysis of the structure-based aggregation propensity computed with A3D. The A3D database implements simple but useful graphical tools for visualizing and interpreting protein structure datasets. It also enables testing the influence of user-selected mutations on protein solubility and stability, all integrated into a user-friendly interface.

Availability and implementation: A3D database is freely available at: http://biocomp.chem.uw.edu.pl/A3D2/hproteome. The data underlying this article are available in the article and in its online supplementary material.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Examples of protein model visualizations from the A3D database. For each database entry, under the Structure tab, two protein copies are presented colored according to (i) the A3D score and (ii) the AlphaFold (AF) model confidence score. The A3D score is visualized in shades from dark blue (highly soluble residues, score < −2.5), through white (no predicted influence on aggregation properties), to dark red (aggregation-prone residues, score > +2.5). The AF per-residue confidence score (pLDDT) is presented in dark blue (very high confidence, pLDDT > 90), light blue (confident, 90 > pLDDT > 70), yellow (low confidence, 70 > pLDDT > 50) and orange (very low confidence, pLDDT < 50). Note that pLDDT < 50 is a reasonably strong predictor of disorder (Tunyasuvunakool et al., 2021), which suggests that a particular region may be unstructured as a linker between domains (see b) or as an inherently disordered domain (see c). (a) An example of a globular protein predicted with high confidence is shown (A color version of this figure appears in the online version of this article)