Bioagents GuideSAFE

🔬 A curated collection of 23,000+ agent skills for empirical research across 8 social science disciplines. | 精选 23,000+ AI Agent 技能库,覆盖8大社会科学学科的实证研究。CoPaper.AI 20分钟完成一篇可复现的规范实证论文,并支持用户上传 Skills。-- Maintained by CoPaper.AI from Stanford REAP.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: bioagents-guide
description: "AI scientist framework for autonomous biological research workflows"
metadata:
  openclaw:
    emoji: "🧬"
    category: "domains"
    subcategory: "biomedical"
    keywords: ["AI scientist", "biological research", "autonomous lab", "drug discovery", "protein design", "bioinformatics"]
    source: "https://github.com/SakanaAI/AI-Scientist"
---

# BioAgents Guide

## Overview

BioAgents -- AI agent systems for biological research -- represent a paradigm shift in how life science experiments are conceived, designed, executed, and analyzed. Building on the foundation of large language models, these systems integrate literature search, hypothesis generation, experimental design, data analysis, and manuscript drafting into semi-autonomous or fully autonomous research pipelines.

The AI Scientist framework (Sakana AI, 2024) demonstrated that language models can conduct end-to-end research: generating ideas, writing code, running experiments, and producing papers. In biology, this approach is being applied to drug discovery, protein engineering, genomics analysis, and systems biology -- domains where the combinatorial complexity of experimental space makes AI-assisted exploration particularly valuable.

This guide covers the architecture of bioagent systems, the biological research tasks they can automate, integration with wet-lab automation, and the methodological considerations for researchers building or evaluating these systems. The focus is on practical patterns that connect AI capabilities to real biological research problems.

## BioAgent Architecture

### System Components

```
BioAgent System Architecture:

┌─────────────────────────────────────────────────┐
│                  ORCHESTRATOR                     │
│  (LLM-based planning and reasoning agent)        │
├──────────┬──────────┬──────────┬────────────────┤
│ LITERATURE│ HYPOTHESIS│ EXPERIMENT│   ANALYSIS    │
│  MODULE   │  MODULE   │  MODULE   │   MODULE      │
├──────────┼──────────┼──────────┼────────────────┤
│ PubMed   │ Causal   │ Protocol │ Statistical    │
│ Semantic │ inference│ generator│ analysis       │
│ Scholar  │ Graph    │ Robot    │ Visualization  │
│ BioRxiv  │ reasoning│ interface│ Interpretation │
│ Patents  │ Novelty  │ LIMS     │ Manuscript     │
│          │ scoring  │ integration│ drafting      │
└──────────┴──────────┴──────────┴────────────────┘
         │              │              │
    ┌────┴────┐   ┌────┴────┐   ┌────┴────┐
    │ Knowledge│   │ Wet Lab  │   │ Compute │
    │ Bases    │   │ Equipment│   │ Cluster │
    └─────────┘   └─────────┘   └─────────┘
```

### Implementing a Literature-Driven Hypothesis Agent

```python
from dataclasses import dataclass
from typing import List, Optional
import json

@dataclass
class Hypothesis:
    statement: str
    mechanism: str
    evidence_for: List[str]
    evidence_against: List[str]
    novelty_score: float
    testability_score: float
    predicted_outcome: str

def generate_hypotheses(
    research_question: str,
    literature_context: List[dict],
    existing_data: Optional[dict] = None,
    n_hypotheses: int = 5,
) -> List[Hypothesis]:
    """
    Generate ranked hypotheses from literature and data context.

    This is a framework for LLM-driven hypothesis generation.
    In practice, the LLM call would go here.
    """
    prompt = f"""
    Based on the following research question and literature context,
    generate {n_hypotheses} testable hypotheses.

    Research question: {research_question}

    Literature findings:
    {json.dumps(literature_context, indent=2)}

    For each hypothesis, provide:
    1. A clear, falsifiable statement
    2. The proposed mechanism
    3. Supporting evidence from the literature
    4. Contradictory evidence
    5. Novelty score (0-1): How novel relative to existing literature
    6. Testability score (0-1): How feasible to test experimentally
    7. Predicted outcome if the hypothesis is correct
    """

    # In production: response = llm.generate(prompt)
    # Parse and return structured hypotheses
    return []  # Placeholder for LLM output parsing

def rank_hypotheses(hypotheses: List[Hypothesis]) -> List[Hypothesis]:
    """Rank hypotheses by composite score (novelty * testability)."""
    for h in hypotheses:
        h.composite_score = h.novelty_score * h.testability_score
    return sorted(hypotheses, key=lambda h: h.composite_score, reverse=True)
```

## Biological Research Tasks for AI Agents

### Drug Discovery Pipeline

```
AI-assisted drug discovery workflow:

1. TARGET IDENTIFICATION
   - Literature mining for disease-gene associations
   - Network analysis of protein-protein interactions
   - Druggability assessment (binding site prediction)
   Tools: OpenTargets, STRING, FPocket

2. HIT IDENTIFICATION
   - Virtual screening of compound libraries
   - De novo molecular generation (SMILES, graph-based)
   - Docking and scoring (molecular dynamics)
   Tools: AutoDock-GPU, RDKit, DeepChem

3. LEAD OPTIMIZATION
   - ADMET property prediction (absorption, distribution, metabolism)
   - Toxicity prediction
   - Multi-objective optimization (potency vs. selectivity vs. ADMET)
   Tools: ADMET-AI, ToxCast, Optuna

4. PRECLINICAL VALIDATION
   - In vitro assay design and analysis
   - Animal model selection and protocol design
   - Pharmacokinetic modeling
   Tools: PK-Sim, literature-based dose prediction
```

### Protein Design and Engineering

```python
# Example: Using ESM-2 embeddings for protein function prediction
# (Practical pattern for bioagent integration)

from transformers import AutoTokenizer, AutoModel
import torch

def get_protein_embeddings(sequences: list, model_name: str = "facebook/esm2_t33_650M_UR50D"):
    """
    Generate protein embeddings using ESM-2 for downstream tasks.
    Applications: function prediction, fitness landscape, design.
    """
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModel.from_pretrained(

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.