TIS transformer is created to annotate translation initiation sites (TISs) on transcripts using nucleotide sequence information. The repository holds the scripts, data, models and model outputs used to perform the benchmarks and remap the human proteome as discussed in NAR Genomics & Bioinformatics.
To apply the TIS Transformer for new transcript sequences of the human transcriptome, check out 
Data files too large to host on GitHub that were created as part of the study, specifically those in the data/, models/, and outputs/ folders, can be downloaded from Zenodo.
Annotations are performed by a machine learning model following a methodology similar to those defined for natural language modelling tasks. Ensemble annotations have been used to obtain transcripts and TIS annotations. The model processes the full transcript sequence to predict the presence of TISs at each position on the transcript. The model architecture is based on that of the Performer, which allows the use of longer input sequences due to the memory efficiency of the attention-based calculations.
The tool has been compared to similar approaches applying TIS prediction based on the transcript nucleotide sequence. More details about the benchmarking approach are listed in the article. The scripts to obtain the scores for TISRover, TITER, and DeepGSR are deposited in scripts/benchmarks. The models are found under models/benchmarks
Using this method, the proteome of the complete human genome has been remapped by training multiple models. Annotations are performed on chromosomes excluded from the training/validation process. The scripts used to train the relevant models, and the models themselves, are stored under /scripts/proteome and /models/proteome
Model predictions are stored under outputs/. For each chromosome, an annotated set of the top 3*(k) predictions have been curated, where (k) denotes the number of translation initiation sites featured by the Ensembl. More information about the column data is given in outputs/README.md.
The annotations performed by the model can be browsed through our custom app. It is furthermore possible to filter the results based on a variety of TIS and transcript properties.
Alternatively, the model outputs for each position on the transcriptome can be acquired under /outputs/
For smaller data sets, check out
pytorch needs to be separately installed by the user. GPU/CUDA support is not required.
Next, the package can be installed running
pip install transcript-transformerDictionary files (YAML/JSON) are the recommended approach to pass arguments to the tool. A genome-level reference and assembly file (*.gtf, *.fa) and output path for the created database file (*.h5) are three required arguments.
For example, using config.yaml:
gtf_path : Homo_sapiens.GRCh38.110.gtf
fa_path : Homo_sapiens.GRCh38.dna.primary_assembly.fa
h5_path : Homo_sapiens.GRCh38.110.h5we can run TIS Transformer:
tis_transformer config.yamlThe tool offers various options to alter the configuration of the model, training parameters and output files. Evaluate these by running tis_transformer --help.
In essence, running tis_transformer will go through the following steps:
- Parse all data to a HDF5 database (h5_path)
- Fine-tune 5 models on non-overlapping folds of the data.
- Get model predictions for all positions of the transcriptome
- Generate table with metadata for the top ranking predictions (determined by a set cut-off)
Various options are available, such as increasing the set of positive predictions provided by TIS Transformer. This can be done by lowering the cut-off with which the positive set is derived:
tis_transformer config.yaml --prob_cutoff 0.01 --resultsNote that --results will prevent the the tool from retraining new models from scratch by skipping steps 1--3 by using the already existing set of predictions of the transcriptome to recreate the final table!
See also https://github.com/TRISTAN-ORF/RiboTIE.
In addition to predicting the full transcriptome as featured in the previous step, it is also possible to use single RNA sequences as input or a path to a .fa file on a trained model. The predict function returns probabilities for all nucleotide positions on the transcript and is stored as a numpy vector format (.npy). When high scoring sites are present, a .csv file containing the relevant positions and additional information is generated.
Six models were trained using different sets of chromosomes. When applying transcript sequences derived from existing genes, ensure the use of the correct model. The model to be used is determined by the chromosome the gene/transcript is located on:
| Model | Chromosomes |
|---|---|
| models/proteome/TIS_transformer_L_1.ckpt | 1, 7, 13, 19 |
| models/proteome/TIS_transformer_L_2.ckpt | 2, 8, 14, 20 |
| models/proteome/TIS_transformer_L_3.ckpt | 3, 9, 15, 21 |
| models/proteome/TIS_transformer_L_4.ckpt | 4, 10, 16, 22 |
| models/proteome/TIS_transformer_L_5.ckpt | 5, 11, 17, X |
| models/proteome/TIS_transformer_L_6.ckpt | 6, 12, 18, Y |
This step ensures the model used is not one trained on related data. For other types of transcript sequences, any model is valid.
Example usage:
transcript_transformer predict AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACGGT RNA models/proteome/TIS_transformer_L_1.ckpt
transcript_transformer predict data/example_data.fa fa models/proteome/TIS_transformer_L_2.ckptMain function arguments, run transcript_transformer predict -h for a complete list:
positional arguments:
input_data path to JSON dict (h5) or fasta file, or RNA sequence
input_type type of input
checkpoint path to checkpoint of trained model
options:
-h, --help show this help message and exit
--prob_th minimum prediction threshold at which site is deemed worthy of attention (default: 0.01)
--save_path save file path (default: results)
--output_type file type of raw model outputs (default: npy)
@article {10.1093/nargab/lqad021,
author = {Clauwaert, Jim and McVey, Zahra and Gupta, Ramneek and Menschaert, Gerben},
title = "{TIS Transformer: remapping the human proteome using deep learning}",
journal = {NAR Genomics and Bioinformatics},
volume = {5},
number = {1},
year = {2023},
month = {03},
issn = {2631-9268},
doi = {10.1093/nargab/lqad021},
url = {https://doi.org/10.1093/nargab/lqad021},
note = {lqad021},
eprint = {https://academic.oup.com/nargab/article-pdf/5/1/lqad021/49418780/lqad021\_supplemental\_file.pdf},
}