Gpnstar Tools#

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

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

Available Tools#

run_gpnstar_vep (Type: RemoteTool)#

Score single-nucleotide variants with GPN-Star, a language model over a multiple-species alignmen…

run_gpnstar_vep tool specification

Tool Information:

  • Name: run_gpnstar_vep

  • Type: RemoteTool

  • Description: Score single-nucleotide variants with GPN-Star, a language model over a multiple-species alignment (human hg38, 100 vertebrates for the primary model). Each variant gets a log-likelihood ratio logit(alt) - logit(ref) at its position, averaged over the forward and reverse-complement windows: negative means the alternate base is less likely than the reference given the alignment, i.e. more likely to be deleterious. Coordinates are 1-based; chromosome names follow the alignment store (1..22, X, Y; a chr prefix is accepted); the reference base must match the alignment’s human base, and a mismatch is refused rather than scored. Returns one score per variant in input order. The alignment store is held by the server operator: unless a full hg38 store has been configured, only the windows of the kit’s eight example variants (chromosomes 1, 2, 3 and 5) can be scored, and any other position is refused with ‘the alignment holds no data’. SNVs only: no indels.

Parameters:

  • variants (array) (required) The variants to score, in the order the scores come back. At most 2048 per call.

  • window_size (integer) (optional) Alignment columns read around each variant (even). The primary model’s native window is 128. Peak GPU memory follows batch_size x window_size, which may not exceed 8192; variants x window_size may not exceed 524288.

  • batch_size (integer) (optional) Variants per forward pass. Does not change the scores’ meaning; it trades memory for speed. batch_size x window_size may not exceed 8192.

  • decimals (integer) (optional) Decimal places for returned scores (default 5).

Example Usage:

query = {
    "name": "run_gpnstar_vep",
    "arguments": {
        "variants": ["item1", "item2"]
    }
}
result = tu.run(query)