Splice sites

Token-level classification that labels each position as a splice acceptor (3' site), splice donor (5' site), or neither. The default model (g0-splice-bigbird) is a G0 BigBird encoder trained on human splice sites from the SpliceAI dataset.

Input

A DNA sequence passed as sequence (inline) or a server-side handle. The model uses a 15,000 bp sliding-window context with a 5,000 bp step and BigBird sparse attention (up to 4,096 tokens). Predictions are most reliable in the central region of each window (positions 5,000 to 10,000 bp); edge positions have less surrounding context.

Output

Per-position acceptor and donor probabilities, returned in the {data, meta} envelope across two tracks (acceptors, donors). Available output formats include probabilities, BED, and GFF3.

Try it

REST API: POST /v1/tasks/splice/predict — see the getting-started guide. MCP: "Fetch the human HBB gene sequence and predict its splice sites." See mcp.md.