SETH is a predictor of nuances of a residue's disorder in proteins. As input, it only needs single protein sequences, which are then encoded in embeddings by the protein language model ProtT5 [1] (Transformer). These embeddings are then passed to a two-layer CNN, whose output are the disorder predictions. The disorder predictions are given as CheZOD scores [2] (or the normalization thereof, if you choose the default output format when executing the script on your local machine), which are continuous values for measuring disorder, where values below 8 signify disorder and values above 8 signify order [2], [3]. If you choose the default output format when executing the script on your local machine, you will additionally receive binary disorder/order predictions (1=disorder, 0=order). Since SETH only needs single protein sequences as input, it can be applied to any protein not exceeding ProtT5's sequence length restrictions and predictions are very fast compared to most disorder predictors achieving high performances.
Precomputed predictions for the human proteome and Swiss-Prot [4] are available at zenodo.
There is a Google Colab script available, where you can execute SETH online to receive disorder predictions (CheZOD scores) and can subsequently download them.
If you have installed python and the libraries specified in the requirements.txt (In case you want to execute SETH on a GPU, which brings significant time advantages, you may need to adjust the torch installation stated in the requirements.txt accordingly (see PyTorch)), the python script SETH_1.py can be executed in your cmd after download by typing SETH_1.py -i <your input fasta file name> -o <the desired name of your output file>.
You may optionally add -f Cs at the end, to only get CheZOD scores for the input protein(s) as output. By default, your output will be a table in CAID format (per protein you get: one colum with the residue number, one colum with the amino acid name, one colum with normalized CheZOD scores, where 1 signifies maximum disorder and 0 maximum order and one colum with binary disorder/order predictions).
When adding the performance of SETH to the large scale evaluation of disorder predictors [3], i.e., evaluating SETH on the same test set, it outperformed all other methods.
In case you are using this method and find it helpful, please cite:
Ilzhöfer D, Heinzinger M and Rost B (2022) SETH predicts nuances of residue disorder from protein embeddings. Front. Bioinform. 2:1019597. https://doi.org/10.3389/fbinf.2022.1019597
[1] Elnaggar, A., Heinzinger, M., Dallago, C., Rihawi, G., Wang, Y., Jones, L., Gibbs, T., Feher, T., Angerer, C., Steinegger, M., Bhowmik, D. & Rost, B. 2021. ProtTrans: Towards Cracking the Language of Life's Code Through Self-Supervised Deep Learning and High Performance Computing. arXiv:2007.06225
[2] Nielsen, J. T. & Mulder, F. A. A. 2020. Quantitative Protein Disorder Assessment Using NMR Chemical Shifts. In: KRAGELUND, B. B. & SKRIVER, K. (eds.) Intrinsically Disordered Proteins. New York, NY: Springer US.
[3] Nielsen, J. T. & Mulder, F. A. A. 2019. Quality and bias of protein disorder predictors. Scientific reports, 9, 5137
[4] The UniProt Consortium 2021. UniProt: the universal protein knowledgebase in 2021. Nucleic Acids Research 49, D480-D489