henhaobio
Sign in

bioRxivBioinformaticsPreprint

Metapredict V2: An update to metapredict, a fast, accurate, and easy-to-use predictor of consensus disorder and structure

Ryan J. EmeneckeriD, Daniel GriffithiD, Alex S. HolehouseiD

PDFDOI

Abstract summary

Claude Opus 5.5 · xhigh effort · 18 Oct 2026 · from the abstract · CC BY-NC-ND 4.0

metapredict V2 claims better disorder-prediction accuracy than the legacy model, shipping as a web server, Python package and command-line tool, illustrated on human transcription factors.

Intrinsically disordered proteins and regions make up 20-40% of most eukaryotic proteomes, and the authors argue that identifying such regions inside large proteins robustly, quickly and easily is critical. This release updates the deep-learning model behind metapredict, becomes the default implementation, and keeps the earlier model available as legacy metapredict.

Key results

  1. ClaimedThe updated model is reported to predict consensus disorder more accurately than the original metapredict, which remains available as the legacy implementation.

    Described as a substantial accuracy improvement over legacy metapredict; the abstract gives no benchmark set, no metric and no comparison against other disorder predictors. · Abstract

  2. ClaimedThe improvement is illustrated on a systematic pass over human transcription factors, and the predictor is reported to behave sensibly on synthetic or non-natural sequences.

    Both are presented as illustrations rather than evaluations: no number of transcription factors, no ground-truth disorder annotation and no accuracy values are reported. · Abstract

  3. ClaimedAccess is through a web server, a Python package and a command-line tool; V2 was released in March 2022 and is now the default metapredict implementation.

    Stated in the key points, which also note the original predictor was released in 2021 and that this manuscript will not be submitted to a journal. · Abstract

Methods, data and reproducibility

An update to a deep-learning predictor of protein disorder: the V2 model supersedes the original as the default implementation, is distributed through a web server, a Python package and a command-line tool, and is compared against the legacy version; it is then applied to human transcription factors and to synthetic, non-natural protein sequences.

  • human transcription factors analysed systematically to illustrate the updated predictor · human
  • synthetic or non-natural protein sequences used as a further check
Methods & tools
metapredict V2, legacy metapredict, Python
Reproducibility gaps
No benchmark set, metric or evaluation code is named for the accuracy comparison between the legacy model and V2.
Limitations (3)
  • The authors describe the manuscript as a summary of how accuracy was improved and a comparison of legacy with V2, and state it will not be submitted to a journal, asking users to cite the original paper instead. · stated by the authors
  • The accuracy gain is unquantified: no held-out benchmark, no metric and no comparison with disorder predictors other than the authors' own earlier model. · noted in this summary
  • The transcription factor and synthetic-sequence analyses are framed as illustrations, so neither supports an error rate for the new model. · noted in this summary

Read this if you annotate intrinsically disordered regions at proteome scale and are deciding whether to switch from legacy metapredict to V2.

Citations
42

Abstract

Intrinsically disordered proteins and protein regions make up 20-40% of most eukaryotic proteomes and play essential roles in a wide gamut of cellular processes, from intracellular trafficking to epigenetic silencing. Given their importance, the ability to robustly, quickly, and easily identify IDRs within large proteins is critical. Here we present metapredict V2, an update to our deep-learning-based disorder predictor metapredict. Metapredict V2 has substantially improved accuracy, more features, and a more user-friendly interface via our web server (https://metapredict.net/), Python package, and command-line tool. To illustrate V2s improved performance we undertake a systematic analysis of human transcription factors, as well as illustrate that metapredict V2 works well for synthetic or non-natural proteins. KEY POINTSO_LIMetapredict is a fast and easy-to-use disorder predictor released in 2021. C_LIO_LIMetapredict V2 was released in March 2022 and includes improved accuracy and new features. C_LIO_LIMetapredict V2 is now the default metapredict implementation, although the original implementation is available as legacy metapredict. C_LIO_LIThis manuscript provides a summary of how we improved the accuracy of metapredict and compares the original version (legacy) to our improved version (V2) C_LIO_LIThis manuscript will not be submitted to a journal; if you use metapredict V2 please cite the original paper and make reference to the fact that V2 is being used. C_LI

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLotthammer '24Emenecker '21Hu '20Redl '23Han '24Necci '21
85% semantic matchBioinformatics
Jeffrey M. Lotthammer, Jorge Hernández‐García, Daniel Griffith+3 more·bioRxiv·Nov 5, 2024· 27 citations