DOCSEARCHbyAlgolia
Submit Docs

ToolUniverse

repository·main·Indexed 23 days ago

https://github.com/mims-harvard/tooluniverse

An ecosystem for building AI scientist systems providing a standardized AI-Tool Interaction Protocol for LLMs to access over 1000 scientific tools, datasets, and models. It features native Model Context Protocol (MCP) server support, a Python SDK for bulk tool calls, and specialized research workflows for biology, chemistry, medicine, and data science. ToolUniverse integrates with Claude Code and Codex, offering capabilities such as cross-validation of claims, literature sweeps, and ID translation across scientific namespaces.

Tokens
358K
Snippets
894
Records
1.4K
Agent score
81%

What's inside tooluniverse

  1. What is ToolUniverse?

    main

    ToolUniverse is an ecosystem designed to create AI scientist systems from any large language model (LLM). It uses the AI-Tool Interaction Protocol to standardize how LLMs identify and call tools.

    Key capabilities include:

    • Universal LLM Support: Works with Claude, GPT, Gemini, Qwen, Deepseek, and open models.
    • Extensive Toolset: Integrates over 1000 machine learning models, datasets, APIs, and scientific packages.
    • MCP Integration: Native Model Context Protocol server support.
    • Async Operations: Support for long-running tasks like protein docking with progress tracking.
    • Compact Mode: A mode that reduces the toolset to 4-5 core discovery tools to save ~99% of the context window.
    • Two-Tier Caching: Uses in-memory LRU and SQLite persistence for high-speed, reproducible tool execution.

  3. Overview of ToolUniverse Research Skills

    main

    ToolUniverse provides a wide array of specialized research skills for AI agents. The core tooluniverse skill acts as a router, automatically dispatching queries to the appropriate specialized skill.

    Key research capabilities include:

    • Genomics & Sequencing: tooluniverse-sequence-retrieval, tooluniverse-variant-analysis, tooluniverse-epigenomics.
    • Drug Discovery & Target Research: tooluniverse-target-research, tooluniverse-drug-research, tooluniverse-binder-discovery, tooluniverse-drug-repurposing.
    • Clinical & Disease Research: tooluniverse-disease-research, tooluniverse-clinical-guidelines, tooluniverse-precision-oncology.
    • Omics Analysis: tooluniverse-rnaseq-deseq2, tooluniverse-single-cell, tooluniverse-proteomics-analysis, tooluniverse-metabolomics-analysis.
    • Safety & Toxicology: tooluniverse-chemical-safety, tooluniverse-pharmacovigilance, tooluniverse-drug-drug-interaction.

    Agents can trigger these by asking natural questions, such as:

    • "Find the E. coli K-12 genome" (triggers tooluniverse-sequence-retrieval)
    • "Assess EGFR as a drug target" (triggers tooluniverse-target-research)

  4. Overview of ToolUniverse Developer Skills

    main

    Developer skills are designed to help build, maintain, and evolve the ToolUniverse ecosystem. They include tools for:

    • Setup & Configuration: setup-tooluniverse, tooluniverse-cs-setup (for Claude Science).
    • Skill Creation: create-tooluniverse-skill, devtu-create-tool, devtu-auto-discover-apis.
    • Maintenance & Optimization: devtu-fix-tool, devtu-optimize-descriptions, devtu-optimize-skills, devtu-docs-quality.
    • Workflow: devtu-github, devtu-self-evolve (for the full self-improvement cycle).

  5. What is ToolUniverse?

    main

    ToolUniverse is a collection of over 1000 scientific tools designed for AI agents. It provides unified access to diverse scientific domains including drug safety data, genomics, literature, and clinical trials.

    You can interact with these tools programmatically using the ToolUniverse class.

    from tooluniverse import ToolUniverse
tu = ToolUniverse()
tu.load_tools()

result = tu.run({
   "name": "UniProt_get_function_by_accession",
   "arguments": {"accession": "P05067"}
})

  6. What is included in ToolUniverse

    main

    The ToolUniverse plugin provides access to over 1000 scientific research tools for biology, chemistry, medicine, and data science. The package includes:

    • MCP server configuration: Enables discovery and execution of ToolUniverse tools.
    • Generated skills: Copies of ToolUniverse skills located in the skills/ directory.
    • Research workflows: Specialized workflows for genomics, drug discovery, clinical analysis, literature review, statistical modeling, and scientific data analysis.

  7. Overview of TxAgent and ToolUniverse Integration

    main

    TxAgent is an AI agent designed for precision therapeutics and personalized treatment recommendations. It can be integrated into the ToolUniverse framework to provide specialized capabilities to AI models.

    Key integration patterns include:

    • Exposing TxAgent as an MCP Server: Using the Model Context Protocol (MCP) to make TxAgent tools accessible via HTTP-based communication.
    • ToolUniverse Integration: Leveraging ToolUniverse's auto-discovery to automatically find and register MCP tools through a unified interface using JSON-based tool definitions.
    • FastMCP Framework: Utilizing the FastMCP Python framework to define tools using decorators and manage transport protocols.

  8. Overview of Translation Tool core files

    main

    The translation toolset consists of three main files:

    • batch_translate_po.py: The primary CLI script for batch translation, supporting single files, directories, status tracking, and configurable parameters.
    • translation_agentic_tool.py: A demonstration script showing a full class implementation of how to use AgenticTool for translation.
    • AGENTIC_TRANSLATION_TOOL_SUMMARY.md: A technical summary containing implementation details, usage examples, and performance analysis.

  9. Validate a variant-effect predictor against DMS data

    main

    Use the tooluniverse-variant-predictor-dms-validation skill to determine if a numeric variant-effect predictor (e.g., AlphaMissense, ESM-C SAE, ESM logits, EVE, or conservation scores) correlates with experimental deep mutational scanning (DMS) measurements on a specific protein.

    Core Functionality

    • Computes per-variant predictor scores.
    • Stratifies variants into 'neutral' vs 'disruptive' groups based on DMS effects.
    • Performs a Mann-Whitney U test on the predictor scores.
    • Executes a parameter sweep (varying K, neutral band, and disruptive quantile) to ensure the statistical results are robust.

    When to use this skill

    • To validate if a predictor is trustworthy for a specific protein of interest.
    • To compare two different predictors on the same DMS dataset (compare per-K Mann-Whitney U p-values and effect sizes).
    • To provide robustness evidence for reviewers or publications.

    When NOT to use this skill

    • For per-position or per-feature interpretation (use tooluniverse-residue-functional-mechanism-interpretation).
    • For single-variant interpretation without DMS data (use tooluniverse-protein-sae-variant-interpretation or tooluniverse-protein-lof-mechanism).
    • For building a new predictor from scratch.

  10. Use the tooluniverse-protein-lof-mechanism skill

    main

    The tooluniverse-protein-lof-mechanism skill is used to propose a specific molecular mechanism for how a missense (coding) variant causes loss-of-function (LoF). It synthesizes evidence from five independent computational layers: AlphaMissense pathogenicity, AlphaFold structural context, ESMC sequence likelihood, SAE feature disruption, and DynaMut2 stability.

    This skill is ideal when you need to know WHY a variant is damaging (e.g., to inform drug rescue strategies or gene therapy decisions) rather than just obtaining a pathogenicity score.

    When NOT to use this skill:

    • For non-coding/regulatory variants (use tooluniverse-variant-to-mechanism).
    • For ACMG pathogenicity classification (use tooluniverse-variant-interpretation).
    • For cancer-specific drivers (use tooluniverse-cancer-variant-interpretation).

  11. Use the Clinical Variant Interpreter skill

    main

    The tooluniverse-variant-interpretation skill provides systematic variant interpretation, moving from raw variant calls to ACMG-classified clinical recommendations. It is designed for:

    • Assessing variant pathogenicity and classification.
    • Annotating VCF data.
    • Performing structural impact analysis for missense variants using AlphaFold2.
    • Predicting functional impacts of non-coding variants (intronic, promoter, UTR, etc.) using deep-learning models like AlphaGenome or Enformer.
    • Producing clinical-grade variant reports for molecular tumor boards.

    Triggers: Use this skill when users ask about variant significance, classification, pathogenicity, or provide VCF data needing annotation.

  12. Explore ToolUniverse Research Skills

    main

    ToolUniverse provides a wide array of specialized AI agent skills designed for advanced biomedical research. These skills are organized into functional domains such as drug discovery, genomics, clinical medicine, and proteomics. Each skill is a standalone package (e.g., tooluniverse-antibody-engineering) that an AI agent can leverage to perform complex, multi-step scientific tasks.

    Available Research Skills by Category:

    Drug Discovery & Pharmacology

    • Adverse Event Detection: Analyze drug signals using FDA FAERS data and disproportionality analysis.
    • Antibody Engineering: Humanization, affinity maturation, and immunogenicity prediction.
    • Binder Discovery: Small molecule binder discovery for protein targets.
    • Chemical Compound Retrieval: Information retrieval from PubChem and ChEMBL.
    • Chemical Safety: Toxicology assessment integrating ADMET-AI and FDA data.
    • Drug Drug Interaction: Prediction of interaction mechanisms and severity.
    • Drug Repurposing: Target-based and disease-driven repurposing strategies.
    • Drug Research: Comprehensive compound research and evidence grading.
    • Drug Target Validation: Computational validation across 10 dimensions.
    • Network Pharmacology: Compound-target-disease network analysis.
    • Target Research: Intelligence gathering across protein info, structure, and pathways.

    Genomics, Epigenomics & Transcriptomics

    • CRISPR Screen Analysis: Functional genomics and synthetic lethal interaction analysis.
    • Epigenomics: Methylation array analysis and CpG detection.
    • Expression Data Retrieval: Omics dataset retrieval from ArrayExpress and BioStudies.
    • Gene Enrichment: Pathway analysis using gseapy, PANTHER, and STRING.
    • GWAS Skills: Includes Fine-Mapping, SNP Interpretation, Study Explorer, Trait-to-Gene, and Drug Discovery.
    • Immune Repertoire Analysis: TCR/BCR sequencing analysis.
    • RNA-seq DESeq2: Differential expression analysis using PyDESeq2.
    • Single Cell: scRNA-seq QC, clustering, and differential expression using scanpy.
    • Spatial Omics & Transcriptomics: Mapping gene expression in tissue architecture.
    • Structural Variant Analysis: Clinical genomics SV classification.
    • Variant Analysis: VCF processing and mutation analysis.
    • Variant Interpretation: Clinical interpretation and ACMG classification.

    Clinical & Precision Medicine

    • Cancer Variant Interpretation: Clinical interpretation of somatic mutations.
    • Clinical Guidelines: Search across 12+ authoritative sources (NICE, WHO, etc.).
    • Clinical Trial Design: Feasibility assessment and regulatory pathway analysis.
    • Clinical Trial Matching: Patient-to-trial matching for precision medicine.
    • Immunotherapy Response Prediction: Predicting ICI response via multi-biomarker integration.
    • Precision Medicine Stratification: Integrating genomic and clinical data for patient grouping.
    • Precision Oncology: Actionable treatment recommendations based on molecular profiles.
    • Rare Disease Diagnosis: Phenotype and genetic-based differential diagnosis.

    Proteomics & Structural Biology

    • Protein Interactions: Network analysis using STRING and BioGRID.
    • Protein Structure Retrieval: Data retrieval from RCSB PDB, PDBe, and AlphaFold.
    • Protein Therapeutic Design: AI-guided de novo design (RFdiffusion, ProteinMPNN).
    • Proteomics Analysis: Mass spectrometry data analysis (quantification, PTMs).
    • Protein Structure Retrieval: Retrieval from PDB, PDBe, and AlphaFold.

    Other Specialized Skills

    • Disease Research: Comprehensive research reports across 10 dimensions.
    • Infectious Disease: Pathogen characterization and drug repurposing.
    • Literature Deep Research: Target disambiguation and theme extraction.
    • Metabolomics: Identification and analysis of metabolites (HMDB, MetaboLights).
    • Multi-Omics Integration: Systems biology integration of multiple datasets.
    • Phylogenetics: Sequence alignment and evolutionary metrics.
    • Polygenic Risk Score: Building and interpreting PRS from GWAS.
    • Statistical Modeling: Regression and survival analysis on biomedical data.