Open-access data: A cornerstone for artificial intelligence approaches to protein structure prediction

Systematic identification of conditionally folded intrinsically disordered regions by AlphaFold2

“…98.5% of the human proteome (Tunyasuvunakool et al 2021). Such remarkable achievements go hand in hand with method developments in structural biology (Eastman et al 2017; Li et al 2013; Punjani et al 2017; Scheres 2012), structural bioinformatics (Andreeva et al 2020; Mariani et al 2013; Mistry et al 2021; Zemla 2003), and protein sequence analysis (Johnson et al 2010; Remmert et al 2011), as well as access to the wealth of information stored in publicly available databases, such as the PDB (Bonvin 2021; Burley & Berman 2021; Burley et al 2019; Goodsell et al 2020) and a multitude of genomic repositories (Howe et al 2021). Indeed, either directly or indirectly, AlphaFold2 relied on the methods and databases listed above to predict structures at the proteome level.…”

“…Instead, IDRs populate an ensemble of interconverting conformations that depends strongly on the primary structure (Das & Pappu 2013), and the properties of these ensembles directly impact on the functions of IDRs (Borcherds et al 2014;Conicella et al 2020;Das et al 2016;Iešmantavičius et al 2014;Kim et al 2021;Maltsev et al 2012;Milles et al 2015;Mittag et al 2008;Sugase et al 2007;Zosel et al 2018). However, experimentally determined structural information for IDR conformational ensembles constitutes only a tiny fraction of the available data for structured proteins (Lazar et al 2021;Varadi et al 2014), and such ensembles are not deposited in the Protein Data Bank (PDB), an online repository that contains more than 150,000 high-resolution structures of biomolecules (Burley & Berman 2021), data which were mined to create AlphaFold2 (Jumper et al 2021a) and RoseTTAFold (Baek et al 2021). Protein structure-prediction programs that make use of available data in the PDB, therefore, will be biased by the relatively few available structures of IDRs, which typically involve those IDRs that fold upon binding to an interaction partner (Smith et al 2021;Wright & Dyson 2009).…”

RCSB Protein Data Bank resources for structure-facilitated design of mRNA vaccines for existing and emerging viral pathogens

“…To make structure-enabled mRNA vaccine design possible, and indeed all structure-enabled science, standardized structure archiving, rigorous validation, expert biocuration, and facile data delivery are essential. Ample evidence of the central role played by the PDB archive has been published in peer-reviewed scientific journals (Burley et al, 2018;Feng et al, 2020;Goodsell et al, 2020;Markosian et al, 2018), going well beyond the fields of structure-guided drug discovery (Burley, 2021;Westbrook and Burley, 2019;Westbrook et al, 2020) and protein structure prediction (Burley and Berman, 2021). The RCSB PDB and its wwPDB partners are dedicated to timely archiving of new results, continuing the 50-year PDB tradition of supporting scientific discovery and technical innovation based on experimental data freely contributed by structural biologists.…”

“…Free availability of rigorously validated and expertly biocurated 3D structures from the PDB archive has enabled progress in the fields of structural biology and structural bioinformatics (reviewed in Burley and Berman, 2021) in myriad ways, including development of new structure-determination methods, structure-guided drug discovery, predicting the impact of point mutations in proteins, comparative or homology protein structure modeling, protein-ligand pose prediction and scoring, prediction of protein-protein interactions, molecular dynamics simulations, and de novo protein structure prediction.…”

Assessing PDB Macromolecular Crystal Structure Confidence at the Individual Amino Acid Residue Level

“…Open access to well-validated, expertly-biocurated PDB structures enables scientific advances across fundamental biology, biomedicine, energy sciences, and biotechnology/bioengineering (Burley et al, 2018; Goodsell et al, 2020; Westbrook and Burley, 2019; Westbrook et al, 2020). PDB structures also played critical roles in efforts aimed at predicting (or computing) atomic-level 3D structure models from protein sequence alone (Burley and Berman, 2021; Burley et al, 2021). Today, AlphaFold2 (Jumper et al, 2021; Tunyasuvunakool et al, 2021) and RoseTTAFold (Baek et al, 2021) support computation of structure models of globular proteins with accuracies comparable to those of lower-resolution experimental methods.…”