CoChat
Sign Up
PreprintCited by 7Protein Structure and DynamicsMachine Learning in Bioinformaticsthermodynamics and calorimetric analyses

The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index.

Homologous Pairs of Low and High Temperature Originating Proteins Spanning the Known Prokaryotic Universe

Open access
Komp, Evan, Alanzi, Humood, Francis, Ryan, Vuong, Chau, Roberts, Logan, Mossallenejad, Amin, Beck, David A. C.·2023·Scientific Data
DOI 10.1101/2023.08.24.554664PMC PMC10560248

Preprint — published as 10.1038/s41597-023-02553-w · Paper page

Open this paper in CoChat — free accountOpen PDF

Abstract

Stability of proteins at high temperature has been a topic of interest for many years, as this attribute is favourable for applications ranging from therapeutics to industrial chemical manufacturing. Our current understanding and methods for designing high-temperature stability into target proteins are inadequate. To drive innovation in this space, we have curated a large dataset, learn2thermDB, of protein-temperature examples, totalling 24 million instances, and paired proteins across temperatures based on homology, yielding 69 million protein pairs - orders of magnitude larger than the current largest. This important step of pairing allows for study of high-temperature stability in a sequence-dependent manner in the big data era. The data pipeline is parameterized and open, allowing it to be tuned by downstream users. We further show that the data contains signal for deep learning. This data offers a new doorway towards thermal stability design models.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1101/2023.08.24.554664

Cites

2 of 2 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 The taxonomic and protein space covered by the protein homologous pairs (N = 69 mil) within the dataset. Left) NCBI taxonomic breakdown of the dataset88. The outer ring depicts super kingdom, Phylum, Class, and Order, where the size of wedges indicates the number of organisms in the classification that contain at least one protein in a learn2therm protein pair. Highly populated Phylum and Class are labelled. The inner ring moving inward is a histogram of the number of proteins participating in pairs per organism followed by a colour mapping labelling organisms as mesophilic in blue and thermophilic in red. Central connections indicate taxa pairs contributing to protein pairs. Right) Two dimensional mapping (using t-SNE) of a sample protein space as determined by Evolutionary Scale Model (ESM) embeddings89. In blue, a sample of data from the ESM Atlas with highest structural confidence. Note that this data contains eukaryotic proteins. In yellow, our proteins in pairs, and in orange, the current largest set of protein pairs across temperature. Size of samples conserves relative size of our proteins vs. the Atlas and reference dataset. For details of the mapping procedure, see Supplementary Information S8.
Fig. 1 The taxonomic and protein space covered by the protein homologous pairs (N = 69 mil) within the dataset. Left) NCBI taxonomic breakdown of the dataset88. The outer ring depicts super kingdom, Phylum, Class, and Order, where the size of wedges indicates the number of organisms in the classification that contain at least one protein in a learn2therm protein pair. Highly populated Phylum and Class are labelled. The inner ring moving inward is a histogram of the number of proteins participating in pairs per organism followed by a colour mapping labelling organisms as mesophilic in blue and thermophilic in red. Central connections indicate taxa pairs contributing to protein pairs. Right) Two dimensional mapping (using t-SNE) of a sample protein space as determined by Evolutionary Scale Model (ESM) embeddings89. In blue, a sample of data from the ESM Atlas with highest structural confidence. Note that this data contains eukaryotic proteins. In yellow, our proteins in pairs, and in orange, the current largest set of protein pairs across temperature. Size of samples conserves relative size of our proteins vs. the Atlas and reference dataset. For details of the mapping procedure, see Supplementary Information S8.CC BY · PubMed Central
Fig. 2 The data distribution as a function of difference in growth temperature between items. top) Histogram of difference in OGT between pairs of organisms. The data is skewed towards 0, but there are still many pairs with temperature differences >30 °C. bottom) Count of protein pairs remaining as minimum OGT difference is increased. We still retain 14 mil protein pairs with OGT difference >30 °C.
Fig. 2 The data distribution as a function of difference in growth temperature between items. top) Histogram of difference in OGT between pairs of organisms. The data is skewed towards 0, but there are still many pairs with temperature differences >30 °C. bottom) Count of protein pairs remaining as minimum OGT difference is increased. We still retain 14 mil protein pairs with OGT difference >30 °C.CC BY · PubMed Central
Fig. 3 Workflow for labelling homologous protein pairs across temperature. Raw data includes RefSeq. 16s rRNA sequences, OGT labels from Engqvist, UniProtKB proteome metadata and proteins. Proteome metadata is parsed to identify a single proteome for highly represented organisms, while retaining data for weakly studied taxa. Proteins are filtered such that only ones from the chosen proteomes and for which we have OGT are kept. Protein pair search space is filtered by first identifying pairs of related organisms via 16s rRNA alignment. Protein pairs are searched for by alignment of sequences. Final database tables are taxa, pairs of meso/thermo taxa, proteins, and pairs of meso/thermo proteins.
Fig. 3 Workflow for labelling homologous protein pairs across temperature. Raw data includes RefSeq. 16s rRNA sequences, OGT labels from Engqvist, UniProtKB proteome metadata and proteins. Proteome metadata is parsed to identify a single proteome for highly represented organisms, while retaining data for weakly studied taxa. Proteins are filtered such that only ones from the chosen proteomes and for which we have OGT are kept. Protein pair search space is filtered by first identifying pairs of related organisms via 16s rRNA alignment. Protein pairs are searched for by alignment of sequences. Final database tables are taxa, pairs of meso/thermo taxa, proteins, and pairs of meso/thermo proteins.CC BY · PubMed Central
Fig. 4 Abbreviated schema for the presented database. The taxa table contains NCBI taxonomic information, from superkingdom to species, as well as 16s rRNA sequence and OGT. The protein table contains protein sequence and external database links. The taxa_pairs and pairs table contain metrics for alignment for 16s rRNA sequences and protein sequences, respectively.
Fig. 4 Abbreviated schema for the presented database. The taxa table contains NCBI taxonomic information, from superkingdom to species, as well as 16s rRNA sequence and OGT. The protein table contains protein sequence and external database links. The taxa_pairs and pairs table contain metrics for alignment for 16s rRNA sequences and protein sequences, respectively.CC BY · PubMed Central
Fig. 5 Melting temperature vs OGT, using data from two third party databases. The black dashed line is identity, and almost all examples fall on the side of TM > OGT. The adage that melting temperature is greater than OGT is clearly supported, with a Spearman’s of 0.85 (P value 0.0), and a binomial test of >99% chance of passing with alternative P-value 2.68e−19.
Fig. 5 Melting temperature vs OGT, using data from two third party databases. The black dashed line is identity, and almost all examples fall on the side of TM > OGT. The adage that melting temperature is greater than OGT is clearly supported, with a Spearman’s of 0.85 (P value 0.0), and a binomial test of >99% chance of passing with alternative P-value 2.68e−19.CC BY · PubMed Central
Fig. 6 Comparison of protein pair quality between our data and Hait et al.’s 1660 protein pairs52. (a) Empirical distribution of local alignment homology as bit score normalised to the average length of both protein strands. Our data has a right shifted score on average with t-test probability = 1.94e−8. (b) Same as A, except with percent identity. Our data has a right shifted score on average with t-test probability = 1.75e−6. (c) Empirical distribution of Jaccard score over Pfam annotations for our data (blue) compared to the baseline data (orange). Our full data has more annotation mismatches on average. When only the 25 mil protein pairs with BLAST coverage >95% are considered, the Pfam annotations become indistinguishable from the baseline with t-test probability = 3.24e−13. (d) Cumulative distribution of FATCAT structural alignment P-value for bins in BLAST coverage uniformly sampled from our data, compared to the baseline structural alignments. Even low coverage pairs are more likely to have less than one in a thousand P-value with binomial confidence >99%.
Fig. 6 Comparison of protein pair quality between our data and Hait et al.’s 1660 protein pairs52. (a) Empirical distribution of local alignment homology as bit score normalised to the average length of both protein strands. Our data has a right shifted score on average with t-test probability = 1.94e−8. (b) Same as A, except with percent identity. Our data has a right shifted score on average with t-test probability = 1.75e−6. (c) Empirical distribution of Jaccard score over Pfam annotations for our data (blue) compared to the baseline data (orange). Our full data has more annotation mismatches on average. When only the 25 mil protein pairs with BLAST coverage >95% are considered, the Pfam annotations become indistinguishable from the baseline with t-test probability = 3.24e−13. (d) Cumulative distribution of FATCAT structural alignment P-value for bins in BLAST coverage uniformly sampled from our data, compared to the baseline structural alignments. Even low coverage pairs are more likely to have less than one in a thousand P-value with binomial confidence >99%.CC BY · PubMed Central
Fig. 7 Predictions on our test set by TemStaPro79. Thermophilic label is > = 60 °C versus < 30 °C. The model makes very few predictions between these two quantities where there is no data, as expected, and is able to recover the actual classes with 91% accuracy.
Fig. 7 Predictions on our test set by TemStaPro79. Thermophilic label is > = 60 °C versus < 30 °C. The model makes very few predictions between these two quantities where there is no data, as expected, and is able to recover the actual classes with 91% accuracy.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 82

  1. 10.1093/protein/gzq025 · DOI
  2. 2020, A thermostable, closed SARS-CoV-2 spike protein trimer, Nat. Struct. Mol. Biol, 27, 934, 10.1038/s41594-020-0478-5 · DOI · Paper page
  3. 2016, Insight into thermophiles and their wide-spectrum applications, 3 Biotech, 6, 81
  4. 10.1007/s11274-015-2005-0 · DOI
  5. 10.1073/pnas.2006753117 · DOI
  6. 10.1016/j.cbpa.2007.01.685 · DOI
  7. 10.1371/journal.pcbi.0030052 · DOI
  8. 10.1039/C6RA16992A · DOI
  9. 2017, Computation of enzyme cold adaptation, Nat. Rev. Chem, 1, 1
  10. 10.1016/j.sbi.2009.08.003 · DOI
  11. 2021, Fundamentals to function: Quantitative and scalable approaches for measuring protein stability, Cell Syst, 12, 547, 10.1016/j.cels.2021.05.009 · DOI
  12. 10.1073/pnas.0503890102 · DOI
  13. 2013, Evolvability of Thermophilic Proteins from Archaea and Bacteria, Biochemistry, 52, 4774, 10.1021/bi400652c · DOI
  14. 10.1016/j.bpj.2011.05.059 · DOI
  15. 10.1073/pnas.1530713100 · DOI
  16. Traxlmayr, M. W. & Shusta, E. V. Directed Evolution of Protein Thermal Stability Using Yeast Surface Display. in Synthetic Antibodies: Methods and Protocols (ed. Tiller, T. ) 45–65 (Springer, 2017). · DOI
  17. 2021, Thermostable cellulases: Structure, catalytic mechanisms, directed evolution and industrial implementations, Renew. Sustain. Energy Rev, 151, 111597, 10.1016/j.rser.2021.111597 · DOI
  18. 10.1093/protein/12.1.47 · DOI
  19. 10.1038/s41592-019-0499-3 · DOI
  20. 10.1002/cbic.201900413 · DOI
  21. 10.1021/acscatal.9b00568 · DOI
  22. 10.1093/protein/gzp090 · DOI
  23. 10.1093/protein/15.3.185 · DOI
  24. 10.1038/nrg3927 · DOI
  25. 10.3390/ijms222111741 · DOI
  26. 10.1038/s41598-018-24760-x · DOI
  27. 2020, A Generative Neural Network for Maximizing Fitness and Diversity of Synthetic DNA and Protein Sequences, Cell Syst, 11, 49, 10.1016/j.cels.2020.05.007 · DOI · Paper page
  28. 2022, Protein design via deep learning, Brief. Bioinform, 23, bbac102, 10.1093/bib/bbac102 · DOI · Paper page
  29. 10.1038/s41586-023-06415-8 · DOI · Paper page
  30. 10.1101/2022.06.03.494563 · DOI · Paper page
  31. 2023, Deep learning of protein sequence design of protein–protein interactions, Bioinformatics, 39, btac733, 10.1093/bioinformatics/btac733 · DOI
  32. 10.1074/jbc.AW119.008144 · DOI
  33. 10.1021/bi902153g · DOI
  34. 10.1038/s41592-020-0848-2 · DOI · Paper page
  35. Casadio, R. , Savojardo, C. , Fariselli, P. , Capriotti, E. & Martelli, P. L. Turning Failures into Applications: The Problem of Protein ΔΔG Prediction. in Data Mining Techniques for the Life Sciences (eds. Carugo, O. & Eisenhaber, F .) 169–185 (Springer US, 2022). · DOI
  36. 2021, Reviewing Challenges of Predicting Protein Melting Temperature Change Upon Mutation Through the Full Analysis of a Highly Detailed Dataset with High-Resolution Structures, Mol. Biotechnol, 63, 863, 10.1007/s12033-021-00349-0 · DOI
  37. 10.1126/science.aah3717 · DOI
  38. 10.1126/science.aai7825 · DOI
  39. 10.1016/0141-8130(82)90049-6 · DOI
  40. 10.1111/febs.13343 · DOI
  41. 2017, Interplay between Protein Thermal Flexibility and Kinetic Stability, Structure, 25, 167, 10.1016/j.str.2016.11.018 · DOI
  42. 10.1038/s41592-020-0801-4 · DOI
  43. 10.1093/nar/gkaa1035 · DOI
  44. 10.1093/nar/gkaa981 · DOI
  45. 10.1021/acssynbio.9b00099 · DOI
  46. 10.1101/2023.03.27.534365 · DOI
  47. 2020, Finding the generalized molecular principles of protein thermal stability, Proteins Struct. Funct. Bioinforma, 88, 788, 10.1002/prot.25866 · DOI
  48. 10.1016/j.sbi.2021.11.001 · DOI
  49. 10.1016/S0301-4622(99)00103-9 · DOI
  50. 2019, Insights on protein thermal stability: a graph representation of molecular interactions, Bioinformatics, 35, 2569, 10.1093/bioinformatics/bty1011 · DOI
  51. 10.1093/protein/gzn001 · DOI
  52. 2022, A Statistical Analysis of the Sequence and Structure of Thermophilic and Non-Thermophilic Proteins, Int. J. Mol. Sci, 23, 10116, 10.3390/ijms231710116 · DOI
  53. 2016, Improved insights into protein thermal stability: from the molecular to the structurome scale, Philos. Trans. R. Soc. Math. Phys. Eng. Sci, 374, 20160141
  54. 10.1016/j.sbi.2016.12.007 · DOI
  55. 2023, DeepSTABp: A Deep Learning Approach for the Prediction of Thermal Protein Stability, Int. J. Mol. Sci, 24, 7444, 10.3390/ijms24087444 · DOI
  56. 10.1101/2020.06.15.153643 · DOI
  57. 10.1101/2022.12.21.521521 · DOI · Paper page
  58. 2021, De novo protein design by deep network hallucination, Nature, 600, 547, 10.1038/s41586-021-04184-w · DOI · Paper page
  59. 2021, Differentiable biology: using deep learning for biophysics-based and data-driven modeling of molecular mechanisms, Nat. Methods, 18, 1169, 10.1038/s41592-021-01283-4 · DOI · Paper page
  60. NCBI Taxonomy: a comprehensive update on curation, resources and tools | Database | Oxford Academic. https://academic.oup.com/database/article/doi/10.1093/database/baaa062/5881509?login=false. · DOI
  61. 10.1126/science.ade2574 · DOI · Paper page
  62. 10.1093/nar/gkv1189 · DOI · Paper page
  63. 2018, Correlating enzyme annotations with a large set of microbial growth temperatures reveals metabolic adaptations to growth at diverse temperatures, BMC Microbiol, 18, 177, 10.1186/s12866-018-1320-7 · DOI
  64. 10.1093/nar/gku989 · DOI · Paper page
  65. 10.1186/1471-2105-10-421 · DOI · Paper page
  66. 10.1186/s12859-016-0992-y · DOI · Paper page
  67. 10.1371/journal.pcbi.1000844 · DOI
  68. 10.1099/ijs.0.059774-0 · DOI
  69. 10.1038/nmeth.3176 · DOI · Paper page
  70. DuckDB | Proceedings of the 2019 International Conference on Management of Data. https://dl.acm.org/doi/abs/10.1145/3299869.3320212. · DOI
  71. 10.1093/NAR/GKAA913 · DOI · Paper page
  72. 10.1093/nar/gkr367 · DOI · Paper page
  73. PyHMMER: a Python library binding to HMMER for efficient sequence analysis | Bioinformatics | Oxford Academic. https://academic.oup.com/bioinformatics/article/39/5/btad214/7131068.
  74. 10.1002/0471250953.bi0301s42 · DOI
  75. 10.1093/nar/gkaa443 · DOI · Paper page
  76. 10.1007/978-1-4939-7000-1_26 · DOI · Paper page
  77. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models | Nucleic Acids Research | Oxford Academic. https://academic.oup.com/nar/article/50/D1/D439/6430488.
  78. 2022, ProTstab2 for Prediction of Protein Thermal Stabilities, Int. J. Mol. Sci, 23, 10798, 10.3390/ijms231810798 · DOI
  79. 2020, Predicting Thermophilic Proteins by Machine Learning, Curr. Bioinforma, 15, 493
  80. 10.3390/ijms24032217 · DOI
  81. 2022, ProtTrans: Toward Understanding the Language of Life Through Self-Supervised Learning, IEEE Trans. Pattern Anal. Mach. Intell, 44, 7112, 10.1109/TPAMI.2021.3095381 · DOI · Paper page
  82. 10.57967/hf/0815 · DOI

Metadata

Authors
Komp, E., Alanzi, H., Francis, R., Vuong, C., Roberts, L., Mossallenejad, A., & Beck, D. A. C.
Affiliations
University of Washington (US)
Corresponding author
  • Komp, Evan
  • Beck, David A. C.
ORCID
Venue
Scientific Data
Publisher
openRxiv
Published
2023-08-26
Volume
10
Issue
1
Pages
682-682
ISSN
2052-4463 (ISSN-L)
DOI
10.1101/2023.08.24.554664
PMC
PMC10560248
OpenAlex
W4387424814
Cited by
7
References
2
Citations by year
2026: 1 · 2025: 5 · 2024: 1
Type
data paper
Language
en
Open access
Licence
cc-by
Locations
  • Scientific Data (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Science Foundation (1934292)
  • University of Washington
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Physical Sciences › Chemistry › Physical and Theoretical Chemistry › thermodynamics and calorimetric analyses
Keywords
protein thermal stability, prokaryotic proteins, protein design, deep learning, protein sequence data
MeSH terms
Prokaryotic Cells, Proteins, Temperature, Protein Stability

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review