Index of every page on this site: llms.txt
All skills
k-dense-ai avatar

/genomic-intelligence

@ec59a29
Run it yourself

Predicts regulatory features, gene structure, and expression directly from DNA sequence using Genomic Intelligence's hosted transformer DNA language models — no local GPU or model weights. Six tasks over a REST API and a hosted MCP server (keyless public demo): promoter regions, splice donor/acceptor sites, enhancer activity, chromatin state, sequence-to-expression (log TPM), and de-novo gene annotation, plus a composite find-genes-then-predict-expression workflow. Use when the user has a gene symbol, a genomic region, or a DNA/FASTA sequence and wants any of these predictions, mentions Genomic Intelligence, genomicintelligence.ai, api.genomicintelligence.ai, or mcp.genomicintelligence.ai.

Use this Skill: https://skilld.dev/gh/k-dense-ai/scientific-agent-skills/genomic-intelligence

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencessequence-acquisition.md

≈1.6k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Sequence acquisition and TSS coordinates

Acquire reference sequence by gene or genomic region, then submit it to GI. Ensembl REST is public; GI REST prediction needs a key. find_genes is a prediction tool, not a sequence acquisition tool.

What the MCP acquisition tools actually retrieve

In reviewed gi-mcp 0.1.0a21, gene metadata first uses a bundled coordinate catalog, then cached/live Ensembl lookup with stale-cache fallback. Sequence bytes use cache, UCSC, Ensembl, then stale cache. The server may therefore answer without a current Ensembl request. Record meta.source, locus_source, assembly, catalog_release, and any stale indicators, along with data.region, strand, and gene/transcript identifiers. Hosted HBB expression acquisition in this review returned a catalog locus and cached GRCh38 bases; a 100-bp region returned UCSC bases. Neither was a live Ensembl lookup.

The default human assembly is GRCh38. Use explicit Ensembl production names for portable non-human requests (for example mus_musculus, drosophila_melanogaster). Some aliases such as human and mouse are accepted; the old blanket claim that mouse is rejected was incorrect. Do not silently substitute assemblies: the GI VCF workflow separately targets GRCh37.

Reference acquisition does not incorporate a sample's variants or edits. Supply validated local sequence for variant-bearing, synthetic or non-reference input. The acquisition layer can admit broader IUPAC bases than the prediction API's A/C/G/T/N alphabet. For raw-base input check before prediction and record any deliberate masking; a handle-only preview cannot establish the whole alphabet. For handles, surface any prediction-time alphabet failure and reacquire the bases if a correction is needed rather than guessing from the preview.

Direct Ensembl REST routes

Purpose Public GET route Important parameters/result
Gene symbol /lookup/symbol/{species}/{symbol} expand=1 includes transcripts; JSON contains gene coordinates, strand and assembly
Stable ID /lookup/id/{id} expand=1 for gene transcripts; verify returned object type/species
Sequence region /sequence/region/{species}/{region} Region 11:5222473..5231670:-1; request Content-Type: text/plain or FASTA and parse exactly one record

Coordinates are 1-based, inclusive; sequence length is end - start + 1. A strand=-1 region returns the reverse complement, so do not reverse it again. Ensembl documents a 10-Mb region maximum; GI still accepts only up to 500,000 bp. Use bounded HTTP timeouts, honor rate-limit responses, and validate status, content type, alphabet and exact length before passing a response as DNA. A successful HTML response is not sequence.

The symbol lookup, ID lookup, and region sequence contracts were source-reviewed. Direct Ensembl documentation and sequence probes returned HTTP 500 during this review; the equivalent hosted acquisition checks succeeded through its cache/UCSC paths. Treat direct-network availability as unverified, not a reason to claim the endpoint is retired.

Expression: select a transcript, then build the window

Expression scores one half-open slice sequence[tss_index - 4599 : tss_index + 4599]. In a 9,198-bp gene-sense window the TSS base is at offset 4,599: 4,599 upstream bases, the TSS base, and 4,598 downstream bases. Its asymmetry reverses in genomic coordinates.

Use the intended transcript, or explicitly choose Ensembl's canonical transcript from the expanded lookup. On plus strand its TSS is transcript start; on minus strand it is transcript end. Gene-body boundaries can differ substantially and are not a substitute. Canonical status is a reference convention, not evidence that this transcript is active in the assayed cell type.

The MCP helper prefers a canonical transcript but falls back to the gene boundary if none is available. Check data.tss_source == "canonical-transcript" or resolve the intended transcript yourself; an exact 9,198-bp length alone does not establish correct TSS placement.

def expression_region(tss, strand):
    """Return 1-based inclusive genomic bounds with gene-sense TSS offset 4599."""
    if not isinstance(tss, int) or isinstance(tss, bool) or tss < 1:
        raise ValueError("tss must be a positive 1-based integer")
    if isinstance(strand, bool) or strand not in (-1, 1):
        raise ValueError("strand must be +1 or -1")
    start, end = ((tss - 4599, tss + 4598) if strand == 1
                  else (tss - 4598, tss + 4599))
    if start < 1:
        raise ValueError("TSS window extends before chromosome start")
    return start, end

# HBB minus-strand canonical TSS in the reviewed GRCh38 reference:
assert expression_region(5227071, -1) == (5222473, 5231670)
# After fetching with strand=-1, 5231670 - 4599 == 5227071.

Also reject a chromosome-end truncation: check the returned length is exactly 9,198 and preserve the transcript accession/version, assembly and fetch region. For a wider gene-sense locus, compute tss_index = tss - start on plus strand or end - tss on minus strand, after stripping sequence whitespace. Require 4599 <= tss_index <= len(sequence) - 4599. Direct expression does not pad or find TSSs. Verify meta.sequence_length, data.input.tss_index, and meta.task_specific_counts.scored_window against those expected values.

Other tasks

Gene-body sequence is appropriate only when it includes the relevant regulatory context. For promoter work fetch upstream/flanking sequence or a TSS-centered region; an unflanked gene body can omit the promoter. For splice, keep transcript orientation and sufficient locus context. For annotation and the composite, plus-strand genomic input gives straightforward coordinate mapping while models can find transcripts on both strands. Match the chosen model's organism and context window (2,000/15,000/249/1,000 bp for current promoter/splice/enhancer/ chromatin defaults); a request above its admission floor can still be padded.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning6d3 checks · Risk SAFE
  • Gen Agent Trust Hub6d

    The skill is safe for its intended purpose of DNA sequence analysis using hosted models. It correctly manages API keys through environment variables and uses standard libraries for networking. The verdict is low because the skill processes genomic data from external sources and user input, which creates a surface for indirect prompt injection, although the risk is mitigated by specific genomic alphabet validation.

  • Socket26d

    No alerts

  • Snyk26d

    Risk: MEDIUM · 1 issue

Signed by skilld at ec59a29. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 2 months ago
Other metadata
compatibility
Python 3.10+ with the `requests` library for the REST examples. Network access required. The REST `/v1` API needs a `GI_API_KEY` (a `gi_` bearer); the hosted MCP server at mcp.genomicintelligence.ai/mcp works keyless against a rate- and concurrency-limited public demo tier, key optional.
metadata
{
  "version": "1.4",
  "last-reviewed": "2026-10-01",
  "skill-author": "Genomic Intelligence",
  "trigger-keywords": "DNA sequence prediction, regulatory genomics, promoter prediction, splice site prediction, enhancer activity, chromatin state, gene expression prediction, sequence to expression, log TPM, gene annotation, transcript prediction, DNA language model, genomic intelligence, hosted inference, Ensembl sequence, FASTA prediction, cis-regulatory, TSS window, DeepSEA, DeepSTARR, BigBird splice, MCP genomics",
  "openclaw": {
    "primaryEnv": "GI_API_KEY",
    "envVars": [
      {
        "name": "GI_API_KEY",
        "required": false,
        "description": "Optional gi_ bearer key for the REST /v1 API and higher MCP rate and concurrency limits. The hosted MCP demo runs keyless; request a key at contact@genomicintelligence.ai."
      }
    ]
  }
}

README badge

README badge for k-dense-ai/scientific-agent-skills/genomic-intelligence