hiperrag

name	hiperrag
description	HPC-scale literature mining and retrieval-augmented generation for structural biology
metadata	{"openclaw":{"requires":{"env":["OPENAI_API_KEY"],"bins":["python3"]},"primaryEnv":"OPENAI_API_KEY"}}

HiPerRAG — Literature Mining

Search and synthesize structural biology literature at scale using HiPerRAG.

Capabilities

• Literature search: Query PubMed, bioRxiv, and local document stores via RAG retrieval
• RAG synthesis: Generate evidence-based summaries from retrieved papers
• Target identification: Identify druggable targets and interacting proteins from literature evidence
• Binder discovery: Find starting scaffolds and binder sequences from clinical/literature evidence
• Evidence ranking: Score and rank evidence quality with hallmark scoring

Prompt Modes

• interactome: Identify interacting proteins, pathways, and therapeutic rationales for a target protein
• binder_design: Find starting binder sequences and scaffolds for BindCraft optimization

Usage

python skills/hiperrag/scripts/run_rag.py \
  --query "Identify interacting proteins for KRAS" \
  --target-protein KRAS \
  --prompt-type interactome \
  --output-dir ./artifacts

Parameters

• query (required): Research question or protein target
• target_protein (required): Target protein name (e.g., KRAS, TP53)
• prompt_type: Prompt mode — interactome or binder_design (default: interactome)
• generator: LLM backend — openai, vllm, or argo (default: openai)
• model: Model name for the generator (default: gpt-4o)
• retrieval_top_k: Number of documents to retrieve (default: 200)
• output_dir: Directory for artifact storage (default: ./artifacts)
• rag_config: Path to a YAML/JSON RAG config file (optional, overrides CLI flags)