Detect splice donor and acceptor sites in FASTA sequences via the Genomic Intelligence API.
npx skills add https://github.com/ClawBio/ClawBio --skill gi-splice
Please help me install this Agent Skill. Skill: gi-splice Source: https://github.com/ClawBio/ClawBio/tree/main/skills/gi-splice Command: npx skills add https://github.com/ClawBio/ClawBio --skill gi-splice
Predicting splice sites normally requires local GPU infrastructure, model weights, and careful preprocessing with tools like SpliceAI. This Skill replaces that setup with a single CLI call to the hosted Genomic Intelligence G0 BigBird transformer, returning ranked donor/acceptor sites with positions and probabilities in about a second.
Run the gi-splice skill with the demo flag to predict splice sites in the bundled HBB gene body and review the generated report.
High-intent search queries and answers about installing and using this skill.
Run the gi-splice CLI with a single-record FASTA file containing the gene body: python skills/gi-splice/gi_splice.py --input my_gene.fa --output report_dir. The skill posts the sequence to the Genomic Intelligence API and returns called donor and acceptor sites with scores.
The skill accepts a single-record FASTA file of 100 to 500,000 bp, typically a full gene body from 5'UTR to 3'UTR including introns. Minus-strand genes must be reverse-complemented to gene-sense orientation before submission, since wrong-strand input still produces plausible-looking output.
Yes, remote inference requires a Genomic Intelligence partner key supplied via the --api-key flag or the GI_API_KEY environment variable. A shared hackathon-tier key ships in .env.example for opt-in demo use, and individual keys can be requested from [email protected].
The skill uploads your FASTA sequence to the hosted Genomic Intelligence API, so identifiable patient data should not be submitted without an appropriate data-use agreement. The skill is intended for research and development use, not clinical or diagnostic decisions.
Sequences outside the 100 to 500,000 bp bounds are rejected with a 422 validation_failed error, and the skill checks length locally before sending a request. Sequences under the model's 15,000 bp context window are accepted but scored against padded context, which can degrade accuracy.
A called site's start and end define a variable-width tokenizer token span of roughly 4 to 10 bp, not an exact junction base. Downstream intersections with VEP consequences or reference exon boundaries should be treated as span-against-position comparisons.
pradyumnasagarCompute reverse complements and complements for DNA/RNA sequences with IUPAC ambiguity support.
NeuroAIHubAnalyze non-coding variant effects on expression, chromatin, and splicing via AlphaGenome API.
K-Dense-AIPredict promoters, splice sites, enhancers, chromatin state, and gene expression from DNA sequences via hosted transformer models.
fgarofalo56Run local AI models via a self-hosted OpenAI-compatible API.
