Skip to content

Repository files navigation

NetStart 2.0: Prediction of Eukaryotic Translation Initiation Sites Using a Protein Language Model

NetStart 2.0 is a deep learning-based model that predicts canonical translation initiation sites in mRNA transcripts in species across the eukaryotic domain.

About

NetStart 2.0 integrates the ESM-2 protein language model for assessing transitions from non-coding to coding regions with local sequence context and taxonomical information.

NetStart 2.0 Online Server

For smaller datasets, the NetStart 2.0 prediction server is available for use here.

Training and test sets

The datasets used for training and testing NetStart 2.0 can be downloaded from the prediction server from the section Data.

Run NetStart 2.0 Locally

Requirements

NetStart 2.0 has been developed in Python, with the following package versions:

  • torch==1.12.1
  • numpy==1.23.5
  • pandas==2.0.3
  • tqdm==4.62.3
  • transformers==4.36.0

Usage Instructions

NetStart 2.0 can be run via the command line by cloning this repository and installing the required packages. To test the installation, run the following command from project root:

python3 ./predict_with_netstart2.py -o chordata -in ./data_example/input_file.fasta

NetStart 2.0 can be run to predict on your own data using the general command:

python3 ./predict_with_netstart2.py [optional arguments] -o ORIGIN -in INPUT_FILENAME 

Please note that the program uses the information in the /src directory.

The sequence origin and input file (in FASTA format) are required. Additionally, NetStart 2.0 accepts a range of optional arguments:

Input Argument Description
-out, --output_filename Name of the output file without the file extension.
--output_results Format of the prediction results. Options are: all (returns predicted probabilities for all ATGs in the input sequence(s)), max_prob (returns only the ATG with the highest predicted probability for each input sequence), and threshold (returns all ATGs with a predicted probability above the specified threshold). Default: all.
--threshold Sets the threshold for filtering predictions. This option is only applicable when --output_results is set to threshold. Default value: 0.625.
--gzip_outfile Specifies whether the output file should be gzipped. Default value: False.
--batch_size Specifies the number of samples to process together in a single pass during prediction. Default value: 64.
-o, --origin Origin of the sequence(s) (refer to the Input Sequence Origin Specifications table for options).
-in, --input_filename Input file in FASTA format. The allowed input alphabet is A, C, G, T, U and N (unknown). All the other letters will be converted to N before processing. T and U are treated as equivalent. The input file can also be provided in gzipped version with a .gz extension.
--compute_device Which hardware accelerator to use. Options are: cuda (NVIDIA GPU), mps (Apple Silicon), or cpu. The program will automatically fall back to CPU if the requested device is unavailable.
--include_reverse_complement Specifies whether NetStart 2.0 should predict on both strands. If you want to use NetStart 2.0 to predict on genomic sequences, you can use this option. Please note that NetStart 2.0 was trained for predicting on transcripts. Performance on genomic sequences may vary and is highly influenced by the presence of introns. We refer to the paper for an assessment of this impact. Default: False.

The predictions are returned as a .csv file, see Output format description for description.

Input Sequence Origin Specifications

Sequence Origin Input Argument (-o)
Chordata chordata
Alligator mississippiensis a_mississippiensis
Anolis carolinensis a_carolinensis
Bos taurus b_taurus
Canis lupus c_lupus
Columba livia c_livia
Danio rerio d_rerio
Equus caballus e_caballus
Gallus gallus g_gallus
Gorilla gorilla g_gorilla
Homo sapiens h_sapiens
Mus musculus m_musculus
Oryctolagus cuniculus o_cuniculus
Oreochromis niloticus o_niloticus
Oryzias latipes o_latipes
Ovis aries o_aries
Pan troglodytes p_troglodytes
Rattus norvegicus r_norvegicus
Sus scrofa s_scrofa
Taeniopygia guttata t_guttata
Xenopus laevis x_laevis
Arthropoda arthropoda
Apis mellifera a_mellifera
Anopheles gambiae a_gambiae
Daphnia carinata d_carinata
Drosophila melanogaster d_melanogaster
Manduca sexta m_sexta
Tribolium castaneum t_castaneum
Nematoda nematoda
Caenorhabditis elegans c_elegans
Placozoa placozoa
Trichoplax adhaerens t_adhaerens
Echinodermata echinodermata
Strongylocentrotus purpuratus s_purpuratus
Apicomplexa apicomplexa
Plasmodium falciparum p_falciparum
Toxoplasma gondii t_gondii
Eimeria maxima e_maxima
Euglenozoa euglenozoa
Trypanosoma brucei t_brucei
Leishmania donovani l_donovani
Evosea evosea
Dictyostelium discoideum d_discoideum
Entamoeba histolytica e_histolytica
Fornicata fornicata
Giardia intestinalis g_intestinalis
Streptophyta streptophyta
Arabidopsis thaliana a_thaliana
Brachypodium distachyon b_distachyon
Glycine max g_max
Hordeum vulgare h_vulgare
Lotus japonicus l_japonicus
Medicago truncatula m_truncatula
Nicotiana tabacum n_tabacum
Oryza sativa o_sativa
Phoenix dactylifera p_dactylifera
Selaginella moellendorffii s_moellendorffii
Setaria viridis s_viridis
Solanum lycopersicum s_lycopersicum
Triticum aestivum t_aestivum
Zea mays z_mays
Ascomycota ascomycota
Aspergillus nidulans a_nidulans
Neurospora crassa n_crassa
Saccharomyces cerevisiae s_cerevisiae
Schizosaccharomyces pombe s_pombe
Basidiomycota basidiomycota
Coprinopsis cinerea c_cinerea
Cryptococcus neoformans c_neoformans
Schizophyllum commune s_commune
Ustilago maydis u_maydis
Mucoromycota mucoromycota
Rhizophagus irregularis r_irregularis
Unknown unknown

Output Format Description

The output is provided as a .csv file which contains the following information attributes for each prediction (ATG):

Attribute Description
origin Specifies the origin of the sequence predicted on (provided by the user).
atg_pos States the position of the ATG predicted upon (corresponds to the A in the codon).
entry_line Specifies the fasta header line of the specific sequence.
preds Provides the predicted probability of the specific ATG being a translation initiation site (in the range [0.0, 1.0]).
stop_codon_position States the position of the first in-frame stop codon relative to the ATG prediction upon (position corresponds to the last position of the stop codon).
peptide_len States the length of the hypothetical peptide.
strand States the strand predicted upon (+ denotes the template strand, - denotes the complement strand).

Citation

Nielsen, L. S., Pedersen, A. G., Winther, O., & Nielsen, H. (2025). NetStart 2.0: prediction of eukaryotic translation initiation sites using a protein language model. BMC bioinformatics, 26(1), 216. https://doi.org/10.1186/s12859-025-06220-2

About

NetStart 2.0: Eukaryotic translation initiation site prediction

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages