AlphaFold 2, but not AlphaFold 3, predicts confident but unrealistic β-solenoid structures for repeat proteins

Computational and Structural Biotechnology Journal, Journal Year: 2025, Volume and Issue: 27, P. 467 - 477

Published: Jan. 1, 2025

AlphaFold 2 (AF2) has revolutionised protein structure prediction but, like any new tool, its performance on specific classes of targets, especially those potentially under-represented in training data, merits attention. Prompted by a highly confident for biologically meaningless, randomly permuted repeat sequence, we assessed AF2 sequences composed perfect repeats random different lengths. frequently folds such into β-solenoids which, while ascribed high confidence, contain unusual and implausible features as internally stacked uncompensated charged residues. A number confidently predicted are other advanced methods intrinsically disordered. The instability some predictions is demonstrated molecular dynamics. Importantly, deep learning-based tools predict structures or with much lower confidence suggesting that alone an unreasonable tendency to but unrealistic sequences. potential implications natural (near-)perfect sequence proteins also explored.

Language: Английский

---

Exploring the diversity of anti-defense systems across prokaryotes, phages and mobile genetic elements

Nucleic Acids Research, Journal Year: 2024, Volume and Issue: 53(1)

Published: Dec. 9, 2024

Abstract The co-evolution of prokaryotes, phages and mobile genetic elements (MGEs) has driven the diversification defense anti-defense systems alike. Anti-defense proteins have diverse functional domains, sequences are typically small, creating a challenge to detect homologs across prokaryotic phage genomes. To date, no tools comprehensively annotate within desired sequence. Here, we developed ‘AntiDefenseFinder’—a free open-source tool web service that detects 156 one or more in any genomic Using this dataset, identified 47 981 distributed prokaryotes their viruses. We found some genes co-localize ‘anti-defense islands’, including Escherichia coli T4 Lambda phages, although many appear standalone. Eighty-nine per cent localize only preferentially MGE. However, >80% anti-Pycsar protein 1 (Apyc1) resides nonmobile regions bacterial Evolutionary analysis biochemical experiments revealed Apyc1 likely originated bacteria regulate cyclic nucleotide (cNMP) signaling, but co-opted overcome cNMP-utilizing defenses. With AntiDefenseFinder tool, hope facilitate identification full repertoire MGEs, discovery new functions deeper understanding host–pathogen arms race.

Language: Английский

---

The protein structurome of Orthornavirae and its dark matter

mBio, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 23, 2024

ABSTRACT Metatranscriptomics is uncovering more and diverse families of viruses with RNA genomes comprising the viral kingdom Orthornavirae in realm Riboviria. Thorough protein annotation comparison are essential to get insights into functions proteins virus evolution. In addition sequence- hmm profile‑based methods, structure adds a powerful tool uncover relationships. We constructed an “structurome” consisting already annotated as well unannotated (“dark matter”) domains encoded genomes. used modeling similarity searches illuminate remaining dark matter hundreds thousands orthornavirus The vast majority showed either “generic” folds, such single α-helices, or no high confidence predictions. Nevertheless, variety lineage-specific globular that were new orthornaviruses general particular identified within proteomic orthornaviruses, including several predicted nucleic acid-binding nucleases. addition, we case exaptation cellular nucleoside monophosphate kinase RNA-binding families. Notwithstanding continuing discovery numerous it appears all conserved large groups have been identified. rest proteome seems be dominated by poorly structured intrinsically disordered ones likely mediate specific virus-host interactions. IMPORTANCE Advanced methods for prediction, AlphaFold2, greatly expand our capability identify infer their evolutionary This particularly pertinent known evolve rapidly result often cannot adequately characterized analysis sequences. performed exhaustive prediction comparative uncharacterized results show consists mostly all-α-helical readily assigned function various interactions between host proteins. great although unexpected represented individual

Language: Английский

---

AlphaFold 2, but not AlphaFold 3, predicts confident but unrealistic beta-solenoid structures for repeat proteins

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 30, 2024

Abstract AlphaFold 2 has revolutionised protein structure prediction but, like any new tool, its performance on specific classes of targets, especially those potentially under- represented in training data, merits attention. Prompted by a highly confident for biologically meaningless, scrambled repeat sequence, we assessed AF2 sequences comprised perfect repeats random different lengths. frequently folds such into β-solenoids which, while ascribed high confidence, contain unusual and implausible features as internally stacked uncompensated charged residues. A number confidently predicted are other advanced methods intrinsically disordered. The instability some predictions is demonstrated Molecular Dynamics. Importantly, Deep Learning-based tools predict structures or with much lower confidence suggesting that alone an unreasonable tendency to but unrealistic sequences. potential implications natural (near-)perfect sequence proteins also explored.

Language: Английский

---