ABI: A Plug-in Interface Framework for AI Agents to Drive Bioinformatics Analysis

ABI (Agent-Bioinformatics Interface) is a plugin-based Python abstraction layer that addresses the problem of standardized interfaces for AI agents to perform complex bioinformatics analysis. It drives analysis through a unified workflow (plan→dry-run→execute→inspect→report), supports plugin extension and native AI integration, has been applied in metatranscriptomics scenarios, uses the MIT open-source license, was released in June 2026, and is actively updated.

Bioinformatics analysis involves multi-step data processing (quality control, alignment, etc.), traditionally relying on manual scripts or workflow engines. With the rise of LLMs, it has become possible for AI agents to perform analysis, but they face challenges such as a large number of tools, complex parameters, and significant environmental differences. The goal of ABI is to provide a standardized interface layer that allows AI to call tools in a unified way while maintaining flexibility.

Core Architecture

• CLI Layer: Built on Typer, supports --output-json for structured data output
• Plugin Registry: Dynamically loads plugins via entry_points
• Plugin Interface: Must implement 6 core methods including load_config and build_plan
• Runtime Backend: Supports local and Nextflow modes

Unified Workflow

1. Plan: Build an execution plan and validate inputs and parameters
2. Dry-Run: Simulate execution to generate traceability records
3. Execute: Actually run the tool and monitor status
4. Inspect: Parse outputs to extract metrics
5. Report: Generate Markdown/HTML reports

ABI plugin development is concise and powerful:

1. Implement a plugin class (with attributes like plugin_id, display_name, and 6 core methods)
2. Register the plugin via pyproject.toml
3. Lazy loading mode: Optional dependencies are imported at the method level, allowing plugin modules to be imported without installing all dependencies, simplifying CI/CD and testing.

ABI provides native support for AI agents:

1. Structured output: All CLI commands support --output-json for easy parsing by agents
2. OpenAI tool description export: Generate OpenAI-compatible tool descriptions via abi export-openai-tools
3. Complete traceability chain: Records inputs, outputs, parameters, and runtime environment for each analysis step, meeting reproducibility requirements.

Application Scenarios

• Built-in metatranscriptomics analysis plugin, supporting the complete workflow from raw sequencing data to functional annotation
• Supports metagenomic plasmid analysis via the autoplasm optional dependency

Team Value

• Standardized interface: Reduces learning and maintenance costs
• Extensibility: New workflows can be quickly integrated via plugins
• AI-ready: Natively supports agent calls, facilitating automated pipelines

Technical highlights and best practices of ABI:

1. Lazy loading design: Improves module loading speed and simplifies test environment configuration 2 Dual-mode backend: Local mode is suitable for development and debugging; Nextflow mode supports large-scale cluster computing
2. Configuration-driven: YAML configuration files define parameters, supporting multi-profile switching (test/production)
3. Test-friendly: Provides the assert_plugin_contract function to verify plugin interface implementation.

ABI represents a new paradigm for integrating bioinformatics and AI agents, retaining analysis flexibility while paving the way for automation and intelligence. As an MIT open-source project, it was just released in June 2026 and is actively updated. With more analysis plugins added, it is expected to become an important bridge connecting AI and bioinformatics.