BioPython Agentic AI

A bioinformatics AI agent inspired by the BioPython library, built with an Orchestrator + Skills architecture using progressive disclosure via the Claude API.

Architecture

The agent uses 3 meta-tools instead of registering skills individually:

Meta-tool	Level	Purpose
list_skills	L1	Lists names and descriptions of all available skills
load_skill	L2	Loads full instructions (SKILL.md)
execute_skill	L3	Executes a skill with parameters

The agent discovers its capabilities dynamically, loading context only when needed (progressive disclosure).

Available Skills

Skill	Description
read_fasta	Reads and parses FASTA files
complementarity	Generates complement and reverse complement strands
transcription	Transcribes DNA to RNA (T -> U)
translation	Translates DNA to protein using the codon table
base_composition	Calculates percentage of A, T, C, G and GC content
mutation_detection	Detects point mutations at specific positions
batch_analysis	Runs any skill across all organisms in a FASTA file

Project Structure

BioPython-Agentic-AI/
├── app/
│   ├── __init__.py
│   ├── orchestrator.py        # Root agent with SkillToolset (3 meta-tools)
│   └── skills/
│       ├── __init__.py        # Progressive disclosure (list/load/execute)
│       ├── batch_analysis/
│       │   ├── SKILL.md
│       │   └── tools.py
│       ├── base_composition/
│       │   ├── SKILL.md
│       │   └── tools.py
│       ├── complementarity/
│       │   ├── SKILL.md
│       │   └── tools.py
│       ├── mutation_detection/
│       │   ├── SKILL.md
│       │   └── tools.py
│       ├── read_fasta/
│       │   ├── SKILL.md
│       │   └── tools.py
│       ├── transcription/
│       │   ├── SKILL.md
│       │   └── tools.py
│       └── translation/
│           ├── SKILL.md
│           └── tools.py
├── bio/                       # Core library (Sequence, FastaOrganism classes)
├── data/                      # Data files (Flaviviridae-genomes.fasta)
├── .env.example               # Environment variables template
├── main.py                    # Entry point
└── README.md

Setup

# 1. Clone the repository
git clone https://github.com/MarceloCorreiaData/BioPython-Agentic-AI.git
cd BioPython-Agentic-AI

# 2. Install dependencies
pip install anthropic python-dotenv

# 3. Configure API key
cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

# 4. Run
python main.py

Usage Examples

You: What skills do you have?
You: Read the FASTA file and tell me how many organisms it contains
You: What is the GC content of the sequence ATCGATCG?
You: Translate ATGATGATG to protein
You: Detect A->G mutation at position 1000 in the first virus
You: Read the FASTA file and determine which organism has the most stable DNA

Execution Flow

User asks a question
    |
    v
Orchestrator sends to Claude with 3 meta-tools
    |
    v
Claude calls list_skills (L1) -> lists available skills
    |
    v
Claude calls load_skill (L2) -> reads skill instructions
    |
    v
Claude calls execute_skill (L3) -> executes with parameters
    |
    v
Result returns to Claude -> formulates final response
    |
    v
Response displayed to user