autonomous-oncology-agent

$npx mdskill add swaruplab/operon/autonomous-oncology-agent

Generates prioritized oncology treatment plans from multimodal patient data.

  • Interprets complex cancer cases combining pathology, genomics, and clinical history.
  • Uses vision models for H&E image analysis and web fetch for guideline retrieval.
  • Synthesizes image and text data to recommend therapies with evidence levels.
  • Outputs a prioritized treatment plan with biomarker predictions within minutes.

SKILL.md

.github/skills/autonomous-oncology-agentView on GitHub ↗
---
name: autonomous-oncology-agent
description: Precision Oncology
keywords:
  - oncology
  - multimodal
  - H&E
  - biomarkers
  - NCCN
measurable_outcome: Generate a prioritized treatment plan with evidence levels and predicted biomarker status (MSI/KRAS) within 5 minutes of data ingest.
license: MIT
metadata:
  author: Nature Cancer 2025
  version: "1.0.0"
compatibility:
  - system: Python 3.9+
allowed-tools:
  - run_shell_command
  - web_fetch
---

# Autonomous Clinical AI Agent (Oncology)

This skill implements the capabilities of the "Autonomous Clinical AI Agent" described in Nature Cancer (2025). It combines Large Language Models (LLMs) for reasoning with specialized vision models for pathology image analysis to support precision oncology decision-making.

## When to Use This Skill

*   **Precision Oncology**: For interpreting complex cancer cases involving pathology, genomics, and clinical history.
*   **Biomarker Detection**: To identify status of key biomarkers (MSI, KRAS, BRAF) from pathology slides (H&E).
*   **Guideline Adherence**: To check treatment plans against NCCN or ASCO guidelines (via OncoKB/PubMed).
*   **Multimodal Synthesis**: When you need to combine image data and text reports.

## Core Capabilities

1.  **Vision Transformer Analysis**: Detects MSI status and key mutations (KRAS, BRAF) directly from H&E images.
2.  **Clinical Reasoning**: Synthesizes patient history, pathology, and genomics to recommend therapies.
3.  **Evidence Retrieval**: Integrates real-time knowledge from OncoKB and PubMed.
4.  **Decision Support**: Provides ranked treatment options with evidence levels.

## Workflow

1.  **Input Processing**:
    *   Text: Clinical notes, pathology reports, genomic panels.
    *   Image: H&E histology slides.
2.  **Analysis**:
    *   Vision model predicts molecular features from slides.
    *   LLM extracts key clinical entities (Stage, Histology, Mutations).
3.  **Reasoning**:
    *   Query OncoKB for actionable mutations.
    *   Match against standard of care guidelines.
4.  **Output**: Generate a comprehensive "Tumor Board" style report.

## Example Usage

**User**: "Review this case of metastatic colorectal cancer. The H&E slide is attached. What is the predicted MSI status and recommended first-line therapy?"

**Agent Action**:
1.  Runs vision model on H&E image -> Output: "MSI-High (Predicted)".
2.  Reads clinical notes -> "Patient is fit, ECOG 0."
3.  Consults Knowledge Base -> "MSI-High CRC responds to Pembrolizumab."
4.  Recommends: "Based on predicted MSI-High status, immunotherapy (Pembrolizumab) is recommended over standard chemotherapy..."

More from swaruplab/operon

SkillDescription
agentd-drug-discoveryUse the AgentD workflow to mine evidence, design molecules, and rank candidates with SAR plus ADMET annotations for early drug discovery tasks.
alphafold-databaseAccess AlphaFold 200M+ AI-predicted protein structures. Retrieve structures by UniProt ID, download PDB/mmCIF files, analyze confidence metrics (pLDDT, PAE), for drug discovery and structural biology.
antibody-design-agentAn advanced agent for de novo antibody design and optimization using state-of-the-art protein language models (MAGE, RFdiffusion).
archrSingle-cell ATAC-seq analysis with ArchR (R). The mature R-based scATAC pipeline — Arrow files, doublet inference, iterative LSI + Harmony, clustering, gene scores, MACS2 peak calling, motif enrichment, chromVAR deviations, footprinting, scRNA-seq integration (label transfer), trajectory analysis (built-in + Monocle3 + Slingshot), and ArchR's interactive genome browser. Sister protocol to snapatac2 — pick this for established R workflows.
automated-bioinformatics-pipelinesPipeline maestro
autonomous-biomedical-agentAutonomous biomedical AI agent framework for executing complex research tasks across genomics, drug discovery, molecular biology, and clinical analysis. Use this skill when conducting multi-step biomedical research including CRISPR screening design, single-cell RNA-seq analysis, ADMET prediction, GWAS interpretation, rare disease diagnosis, or lab protocol optimization. Leverages LLM reasoning with code execution and integrated biomedical databases.
basecallingConvert raw Nanopore signal data (FAST5/POD5) to nucleotide sequences using Dorado basecaller. Covers model selection, GPU acceleration, modified base detection, and quality filtering. Use when processing raw Nanopore data before alignment. Note: Guppy is deprecated; use Dorado for all new analyses.
batch-processingProcess multiple sequence files in batch using Biopython. Use when working with many files, merging/splitting sequences, or automating file operations across directories.
bindingdb-databaseQuery BindingDB for measured drug-target binding affinities (Ki, Kd, IC50, EC50). Search by target (UniProt ID), compound (SMILES/name), or pathogen. Essential for drug discovery, lead optimization, polypharmacology analysis, and structure-activity relationship (SAR) studies.
biological-systems-reasoning|