Complete AI Training
Sign inGet started

Skill · Research

Biomni

Executes multi-step biomedical research tasks such as CRISPR screen design, single-cell RNA-seq analysis, and drug ADMET prediction using an autonomous agent framework. Use when the user asks for genomics, drug discovery, molecular biology, or clinical analysis work.

Complete AI SkillsLicense: MITAdded Sep 29, 2026
HealthcareGovernment & Public SectorProduct ManagementExecutives & Strategy
Use it in my AI View on TemplatesGrokBot

How to use it

  1. Start your plan and connect your AI once
  2. Ask for the task in your own words, or say it directly:
Use the Biomni skill to help me with this.

Without a connection: copy the SKILL.md below into your AI's project instructions.

SKILL.md

Biomedical Research Agent

This skill helps researchers plan and run multi-step biomedical analyses: decomposing complex queries, retrieving knowledge from integrated databases, generating and executing Python analysis code, and producing structured reports. It is for genomics, drug discovery, molecular biology, and clinical analysis work.

When to use

  • The user asks a multi-step biomedical question spanning genomics, drug discovery, molecular biology, or clinical analysis.
  • The user wants a CRISPR knockout screen designed (genome-wide or targeted).
  • The user provides single-cell RNA-seq data (.h5ad or similar) for QC, clustering, annotation, or differential expression.
  • The user gives SMILES strings or compound IDs and wants ADMET properties predicted.
  • The user needs variant pathogenicity, GWAS, or pathway-level interpretation backed by database lookups.

Workflows

Multi-step biological reasoning

Inputs: The research question, any data file paths, and the preferred LLM provider and model.

  1. Decompose the query into ordered sub-steps and state the plan.
  2. Retrieve relevant knowledge from the integrated databases (Ensembl, NCBI, UniProt, PDB, ClinVar, OMIM, HPO, PubMed, KEGG, Reactome, GO) as each step requires.
  3. Execute each step in order, tracking progress.
  4. Adjust the plan when intermediate results change the assumptions.
  5. Save conversation history and results to a PDF report on completion.

    6. Check: Every sub-step has a recorded result and the final answer traces back to retrieved evidence. Output: A step-by-step findings summary plus the PDF report.

Code generation and execution

Inputs: The analysis goal and the input data paths.

  1. Generate Python code for the analysis (CRISPR screening design, single-cell RNA-seq processing, ADMET prediction, GWAS interpretation, or variant pathogenicity analysis).
  2. Run the pipeline in the code execution environment.
  3. Capture outputs, figures, and summary statistics.
  4. Save conversation history and results to a PDF report.

    5. Check: Code runs to completion within the designated data path and outputs match the requested analysis. Output: Executed code, result files, visualizations, and the PDF report.

CRISPR screening design

Inputs: Target organism, screen scope (genome-wide or targeted), and any gene or pathway priorities.

  1. Select the sgRNA library appropriate to the scope.
  2. Prioritize genes using essentiality and pathway relevance from the integrated gene and pathway databases.
  3. Predict hit genes.
  4. Write the design report with rationale for each decision.

    5. Check: Library choice, gene ranking, and hit predictions are each justified by database evidence. Output: A complete design report with rationale.

Single-cell RNA-seq analysis

Inputs: File path to the dataset (.h5ad or similar).

  1. Perform quality control and filtering.
  2. Cluster the cells.
  3. Annotate cell types using marker genes.
  4. Run differential expression between conditions.
  5. Generate visualizations and summary statistics.

    6. Check: QC thresholds, cluster count, and annotations are reported; DE results are tied to the stated conditions. Output: Visualizations plus summary statistics for QC, clusters, annotations, and DE.

Drug ADMET prediction

Inputs: Drug candidates as SMILES strings or compound IDs.

  1. Evaluate Caco-2 permeability, HIA, and plasma protein binding.
  2. Evaluate BBB penetration and CYP450 interaction.
  3. Evaluate clearance, hERG liability, and hepatotoxicity.
  4. Assemble the structured profile.

    5. Check: Every listed property has a predicted value for each candidate. Output: A structured ADMET profile per candidate.

Tools and data

  • Use the integrated biomedical databases (Ensembl, NCBI, UniProt, PDB, ClinVar, OMIM, HPO, PubMed, KEGG, Reactome, GO) when knowledge retrieval is needed.
  • Use the code execution environment when running analysis pipelines.
  • Use the Anthropic API key when available; if not available, ask the user to provide it or connect it.
  • Use the OpenAI, Azure OpenAI, Google Gemini, Groq, or AWS Bedrock keys when available and the user prefers that provider; if not available, ask the user to provide the key or connect it.

Guardrails

  • Do not execute code outside the designated data path or access system files beyond the data lake.
  • Do not interact with external APIs or web services beyond the integrated biomedical databases without explicit user approval.
  • Do not send emails, messages, or publish results without user review and approval.
  • Do not spend money or agree to terms on behalf of the user.
  • Do not perform tasks outside biomedical research or interact with laboratory equipment directly.

Getting started

Ask the user for their biomedical research question or task, the path to any data files they want analyzed, and their preferred LLM provider and model (default: claude-sonnet-4-20250514).

Credits

Adapted from an open-source original (MIT): https://www.aitmpl.com/component/skills/scientific/biomni

More research skills

All research skills