Carbon A Tools#

Configuration File: remote_tools/carbon_a_tools.json Tool Type: Remote Tools Count: 1

This page contains all tools defined in the carbon_a_tools.json configuration file.

Available Tools#

carbon_a_annotate_cds (Type: RemoteTool)#

Annotate protein-coding genes in a eukaryotic DNA sequence with Carbon-A (HuggingFaceBio/Carbon-A…

carbon_a_annotate_cds tool specification

Tool Information:

  • Name: carbon_a_annotate_cds

  • Type: RemoteTool

  • Description: Annotate protein-coding genes in a eukaryotic DNA sequence with Carbon-A (HuggingFaceBio/Carbon-A-1.2B), a 1.2B-parameter DNA model that predicts coding sequence at single-base resolution on both strands, followed by its annotation pipeline: a transition network, a Viterbi decode over a gene grammar (start codon to in-frame stop, GT-AG/GC-AG/AT-AC introns, reading frame continued across introns), and a confidence network that scores each gene. Returns gene coordinates, strand, exon count, per-gene confidence and optionally the GFF3 and translated proteins. Give either an inline sequence of at least 100,000 bp or a FASTA file in the provider’s input directory. Needs a provider with a CUDA GPU and runs at roughly 10 kbp per second. Use for: finding genes in a newly assembled eukaryotic genome, re-annotating a region, or scoring how confidently a locus codes for protein.

Parameters:

  • sequence (string) (optional) Genomic DNA to annotate, as bare IUPAC nucleotides with no FASTA header. Carbon-A reads 98,304 bp windows and the pipeline skips records shorter than min_record_length, so this must be at least 100,000 bp unless you lower that deliberately. Pass input_file instead for a whole genome.

  • sequence_id (string) (optional) Name for the inline record, used as the GFF seqid. Default ‘query’.

  • input_file (string) (optional) FASTA or GenBank file name relative to the provider’s Carbon-A input directory, for example ‘yeast/GCF_000146045.2.fna’, optionally gzip-compressed. Absolute paths, ‘..’ segments and URLs are rejected.

  • genetic_code (string) (optional) Translation table used to close genes at an in-frame stop. ‘standard’ is the default; ‘tetrahymena_table6’ is NCBI table 6, which Tetrahymena thermophila (GCF_000189635.1) needs. A code that does not fit the sequence makes the pipeline flag the run, and that flag is returned in warnings.

  • min_confidence (number) (optional) Only return genes whose confidence reaches this. Confidence is the model’s own probability that a gene’s structure is exactly right. 0.1 is the threshold the model card reports its gene counts at. Genes the pipeline gave no confidence are always returned.

  • min_record_length (integer) (optional) Records shorter than this are skipped by the pipeline rather than annotated. Its own floor is 100,000 bp, which is the default here; below that the model is outside its reported range.

  • attn_implementation (string) (optional) Attention kernel. flash_attention_2 produced every result the model card reports and needs the flash-attn package on the provider; sdpa needs no extra package and is not bit-identical. Omit to use the provider’s configured default.

  • max_genes (integer) (optional) Maximum gene records to return, in genome order. The counts reported alongside them cover every gene, not just the returned ones.

  • return_gff (boolean) (optional) Include the full prediction as GFF3 text.

  • return_proteins (boolean) (optional) Include each returned gene’s translated protein sequence.

  • timeout_seconds (integer) (optional) Maximum time for the call, including any wait for another annotation to finish on the provider. Carbon-A runs at roughly 10 kbp per second on one GPU.

  • dry_run (boolean) (optional) Validate the arguments, including that input_file exists on the provider, and return the command that would run without running Carbon-A.

Example Usage:

query = {
    "name": "carbon_a_annotate_cds",
    "arguments": {
    }
}
result = tu.run(query)