CoChat
Sign Up
PreprintCited by 26RNA and protein synthesis mechanismsGenomics and Phylogenetic StudiesMachine Learning in Bioinformatics

The DOI matches its Crossref record; 1 of the 1 references it cites were found in a registry or an index.

TIS Transformer: Remapping the Human Proteome Using Deep Learning

Open access
Clauwaert, Jim, Gupta, Ramneek, McVey, Zahra, Menschaert, Gerben·2021·NAR Genomics and Bioinformatics
DOI 10.1101/2021.11.18.468957PMC PMC9985340

Preprint — published as 10.1093/nargab/lqad021 · Paper page

Open this paper in CoChat — free accountOpen PDF

Abstract

A bstract: The correct mapping of the proteome is an important step towards advancing our understanding of biological systems and cellular mechanisms. Methods that provide better mappings can fuel important processes such as drug discovery and disease understanding. Currently, true determination of translation initiation sites is primarily achieved by in vivo experiments. Here we propose TIS Transformer, a deep learning model for the determination of translation start sites solely utilizing the information embedded in the transcript nucleotide sequence. The method is built upon deep learning techniques first designed for natural language processing. We prove this approach to be best suited for learning the semantics of translation, outperforming previous approaches by a large margin. We demonstrate that limitations in the model performance is primarily due to the presence of low quality annotations against which the model is evaluated against. Advantages of the method are its ability to detect key features of the translation process and multiple coding sequences on a transcript. These include micropeptides encoded by short Open Reading Frames, either alongside a canonical coding sequence or within long non-coding RNAs. To demonstrate the use of our methods, we applied TIS Transformer to remap the full human proteome.

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 · 10.1101/2021.11.18.468957

Cites

1 of 1 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. Schematic of the data and model set-up. (Left) The Ensembl annotation (version 107) is used to determine transcript sequences and translation initiation sites (TISs). Transcripts are grouped by chromosome to create a training, validation and test set. (Right) The performer model allows processing of full transcript sequences, evaluating data through the layers in parallel to obtain model outputs at each position. The model architecture can handle varying input lengths, as identical model weights are applied to transform the data. Through self-attention, sequential information from any site on the transcript can be queried by the model to determine the presence of TISs at any position.
Figure 1. Schematic of the data and model set-up. (Left) The Ensembl annotation (version 107) is used to determine transcript sequences and translation initiation sites (TISs). Transcripts are grouped by chromosome to create a training, validation and test set. (Right) The performer model allows processing of full transcript sequences, evaluating data through the layers in parallel to obtain model outputs at each position. The model architecture can handle varying input lengths, as identical model weights are applied to transform the data. Through self-attention, sequential information from any site on the transcript can be queried by the model to determine the presence of TISs at any position.CC BY · PubMed Central
Figure 2. Model performances on predicting translation initiation sites. (A) (top) A histogram of the model output predictions when evaluated on chromosome 2 (test set). Transcript sites are divided in a negative (blue) and positive (orange) set according to the annotations provided by Ensembl. The dotted line represents the threshold that an equal number of positive model predictions as provided by Ensembl. (bottom) The resulting coding sequences of predicted TISs were evaluated against UniProt using pBLAST. Shown are the fraction of coding sequences returning a good match in relation to the model output score. Only TISs not annotated by Ensembl (i.e., negative set) were considered. (B) Model performances were binned according to transcripts properties. PR AUC performances (left) and the ratio of positive samples in each group (right) are obtained by binning the transcripts according to transcript support level and other properties given to the annotated translation initiation site or transcript (if any, otherwise nan).
Figure 2. Model performances on predicting translation initiation sites. (A) (top) A histogram of the model output predictions when evaluated on chromosome 2 (test set). Transcript sites are divided in a negative (blue) and positive (orange) set according to the annotations provided by Ensembl. The dotted line represents the threshold that an equal number of positive model predictions as provided by Ensembl. (bottom) The resulting coding sequences of predicted TISs were evaluated against UniProt using pBLAST. Shown are the fraction of coding sequences returning a good match in relation to the model output score. Only TISs not annotated by Ensembl (i.e., negative set) were considered. (B) Model performances were binned according to transcripts properties. PR AUC performances (left) and the ratio of positive samples in each group (right) are obtained by binning the transcripts according to transcript support level and other properties given to the annotated translation initiation site or transcript (if any, otherwise nan).CC BY · PubMed Central
Figure 3. Attribution scores of input nucleotides which reflect the importance of each nucleotide towards predicting the annotated TIS. (A) Scores shown are given by the model for the translation initiation site on position 257 of FGFR3 (ENST00000340107) and at position 52 of CYP4A11 (ENST00000475477). (B) Attribution scores for the positions surrounding the translation initiation site (TIS) averaged for all top ranked predictions. a rolling average is given for the three reading frames.
Figure 3. Attribution scores of input nucleotides which reflect the importance of each nucleotide towards predicting the annotated TIS. (A) Scores shown are given by the model for the translation initiation site on position 257 of FGFR3 (ENST00000340107) and at position 52 of CYP4A11 (ENST00000475477). (B) Attribution scores for the positions surrounding the translation initiation site (TIS) averaged for all top ranked predictions. a rolling average is given for the three reading frames.CC BY · PubMed Central
Figure 4. Example transcripts with predicted micropeptide expression. Shown are the model outputs (y-axis) for each position of the transcript (x-axis). For high TIS predictions, the bounds (striped line) and area of the resulting CDSs are shown. Given are the predictions for the transcripts. (A) MKKS (ENST00000652676), (B) SPAAR (ENST00000443779), (C) MTLN (ENST00000426713) and (D) ATF4 (ENST00000676346).
Figure 4. Example transcripts with predicted micropeptide expression. Shown are the model outputs (y-axis) for each position of the transcript (x-axis). For high TIS predictions, the bounds (striped line) and area of the resulting CDSs are shown. Given are the predictions for the transcripts. (A) MKKS (ENST00000652676), (B) SPAAR (ENST00000443779), (C) MTLN (ENST00000426713) and (D) ATF4 (ENST00000676346).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: 48

  1. Aken BL et al. 2016. The Ensembl Gene Annotation System. Database. 2016: · DOI · Paper page
  2. 2013, Translational Repression of the McKusickKaufman Syndrome Transcript by Unique Upstream Open Reading Frames Encoding Mitochondrial Proteins with Alternative Polyadenylation Sites, Biochimica et Biophysica Acta (BBA) - General Subjects, 1830, 2728, 10.1016/j.bbagen.2012.12.010 · DOI
  3. 10.1038/nrg3520 · DOI
  4. 10.1016/j.ab.2014.06.022 · DOI
  5. Cheng J , Dong L , and Lapata M. 2016. Long Short-Term Memory-Networks for Machine Reading. arXiv:1601.06733 [cs]. · DOI
  6. Choromanski K et al. 2021. Rethinking Attention with Performers. arXiv:2009.14794 [cs, stat]. · arXiv
  7. 10.1038/369371a0 · DOI
  8. 10.1016/j.molcel.2015.11.013 · DOI
  9. 2006, The Abundance of Short Proteins in the Mammalian Proteome, PLOS Genetics, 2, e52, 10.1371/journal.pgen.0020052 · DOI
  10. 10.1016/j.heliyon.2020.e04825 · DOI
  11. Jorgensen RA and Dorantes-Acosta AE. 2012. Conserved Peptide Upstream Open Reading Frames Are Associated with Regulatory Genes in Angiosperms. Frontiers in Plant Science. 0: · DOI
  12. 10.1016/j.compbiomed.2015.09.010 · DOI
  13. 2019, DeepGSR: An Optimized Deep-Learning Structure for the Recognition of Genomic Signals and Regions, Bioinformatics, 35, 1125, 10.1093/bioinformatics/bty752 · DOI
  14. 2018, MOXI Is a Mitochondrial Micropeptide That Enhances Fatty Acid β-Oxidation, Cell Reports, 23, 3701, 10.1016/j.celrep.2018.05.058 · DOI
  15. 10.1038/nature21034 · DOI
  16. 10.1038/s41587-022-01369-0 · DOI
  17. Mudge JM , Ruiz-Orera J , Prensner JR , Brunet MA , Gonzalez JM , et al. 2021. A Community-Driven Roadmap to Advance Research on Translated Open Reading Frames Detected by Ribo-Seq. bioRxiv. 2021.06.10.447896. · DOI
  18. Parikh A , Täckström O , Das D , and Uszkoreit J 2016. A Decomposable Attention Model for Natural Language Inference. In: Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Austin, Texas: Association for Computational Linguistics, pp. 2249–2255. · DOI
  19. 10.1002/bies.201400103 · DOI
  20. 2021, Most Non-Canonical Proteins Uniquely Populate the Proteome or Immunopeptidome, Cell Reports, 34, 108815, 10.1016/j.celrep.2021.108815 · DOI
  21. 10.1093/bioinformatics/btm177 · DOI
  22. 2018, Mitoregulin: A lncRNA-Encoded Microprotein That Supports Mitochondrial Supercomplexes and Respiratory Efficiency, Cell Reports, 23, 3710, 10.1016/j.celrep.2018.06.002 · DOI
  23. Sundararajan M , Taly A , and Yan Q. 2017. Axiomatic Attribution for Deep Networks. arXiv:1703.01365 [cs]. · arXiv
  24. 10.2527/jas2016.94supplement4184x · DOI · Paper page
  25. 2017, Attention Is All You Need, Advances in Neural Information Processing Systems, 30, 5998
  26. 2021, The Role of Micropeptides in Biology, Cellular and Molecular Life Sciences, 78, 3285, 10.1007/s00018-020-03740-3 · DOI
  27. Wang S , Li BZ , Khabsa M , Fang H , and Ma H. 2020. Linformer: Self-Attention with Linear Complexity. arXiv:2006.04768 [cs, stat]. · arXiv
  28. 10.1016/j.dsp.2021.103202 · DOI
  29. 10.1016/S0968-0004(03)00051-3 · DOI
  30. Xiong Y , Zeng Z , Chakraborty R , Tan M , Fung G , Li Y , and Singh V. 2021. Nystr\”omformer: A Nystr\”om-Based Algorithm for Approximating Self-Attention. arXiv:2102.03902 [cs]. · DOI · Paper page
  31. 10.1093/nar/gkv1157 · DOI
  32. 2020, Emerging Role of Long Noncoding RNA-Encoded Micropeptides in Cancer, Cancer Cell International, 20, 506, 10.1186/s12935-020-01589-x · DOI
  33. 10.1074/jbc.R116.733899 · DOI
  34. 2020, Advances in Neural Information Processing Systems, 33, 17283
  35. 10.1093/bioinformatics/btx247 · DOI
  36. 10.1093/bioinformatics/16.9.799 · DOI
  37. 10.1504/IJDMB.2018.094781 · DOI
  38. 10.1016/j.ab.2014.06.022 · DOI
  39. K. Choromanski , V. Likhosherstov , D. Dohan , X. Song , A. Gane , T. Sarlos , P. Hawkins , J. Davis , A. Mo-hiuddin , L. Kaiser , D. Belanger , L. Colwell , and A. Weller . Rethinking Attention with Performers. arXiv:2009.14794 [cs, stat], Mar. 2021. · arXiv
  40. 10.1016/j.heliyon.2020.e04825 · DOI
  41. 10.1016/j.compbiomed.2015.09.010 · DOI
  42. 10.1093/bioinformatics/bty752 · DOI
  43. 10.1038/s41587-022-01369-0 · DOI
  44. 10.1093/bioinformatics/btm177 · DOI
  45. 10.1016/j.dsp.2021.103202 · DOI
  46. 10.1093/bioinformatics/btx247 · DOI
  47. 10.1093/bioinformatics/16.9.799 · DOI
  48. 10.1504/IJDMB.2018.094781 · DOI

Metadata

Authors
Clauwaert, J., Gupta, R., McVey, Z., & Menschaert, G.
Affiliations
  • Ghent University (BE)
  • Novo Nordisk (United Kingdom) (GB)
Corresponding author
  • Clauwaert, Jim
  • Menschaert, Gerben
ORCID
Venue
NAR Genomics and Bioinformatics
Publisher
openRxiv
Published
2021-11-19
Volume
5
Issue
1
Pages
lqad021-lqad021
ISSN
2631-9268 (ISSN-L)
DOI
10.1101/2021.11.18.468957
PMC
PMC9985340
OpenAlex
W4323076266
Cited by
26
References
1
Citations by year
2026: 8 · 2025: 6 · 2024: 3 · 2023: 9
FWCI
2.32
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • NAR Genomics and Bioinformatics (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Ghent University Academic Bibliography (Ghent University) (publishedVersion)
  • Ghent University Academic Bibliography (Ghent University) (submittedVersion, cc-by)
Funders
Universiteit Gent
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › RNA and protein synthesis mechanisms
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
Keywords
translation initiation sites, transformer models, transcript sequences, micropeptides, open reading frames, long non-coding RNA, alternative translation initiation, annotation quality

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