Pan-Cancer Multi-Omics Agent

The Pan-Cancer Multi-Omics Agent integrates multi-omics data across cancer types to identify shared oncogenic drivers, discover novel subtypes, and enable cross-cancer therapeutic insights. It leverages TCGA, CPTAC, and other pan-cancer resources with deep learning for comprehensive cancer characterization.

When to Use This Skill

• When analyzing patient tumors in context of pan-cancer molecular profiles.
• To identify shared drivers and vulnerabilities across cancer types.
• For discovering novel molecular subtypes that span histological boundaries.
• When prioritizing therapeutic targets with pan-cancer evidence.
• To benchmark single-cancer findings against pan-cancer patterns.

Core Capabilities

1. Pan-Cancer Subtyping: ML-based clustering across 32+ cancer types to identify molecular subtypes transcending tissue of origin.

2. Driver Discovery: Integrate mutation, expression, and CNV data to identify oncogenic drivers using pan-cancer statistical power.

3. Multi-Omics Fusion: Deep learning integration of mRNA, miRNA, methylation, and protein data for comprehensive profiles.

4. Pathway Analysis: Identify dysregulated pathways with pan-cancer prevalence and therapeutic implications.

5. Survival Modeling: PRISM framework for multi-omics prognostic marker discovery and survival prediction.

6. Therapeutic Matching: Map patient profiles to pan-cancer drug sensitivity data and clinical trial evidence.

TCGA Pan-Cancer Atlas Integration

Data Type	Samples	Application
Somatic mutations	11,000+	Driver identification
Copy number	11,000+	Amplifications/deletions
mRNA expression	11,000+	Expression subtypes
miRNA expression	10,000+	Regulatory networks
DNA methylation	10,000+	Epigenetic subtypes
Protein (RPPA)	8,000+	Pathway activation

Workflow

1. Input: Patient multi-omics data (mutations, CNV, expression, methylation).

2. Normalization: Harmonize data to TCGA reference standards.

3. Classification: Assign to pan-cancer molecular subtypes.

4. Driver Analysis: Identify patient-specific drivers in pan-cancer context.

5. Pathway Scoring: Calculate pathway activation scores.

6. Therapeutic Matching: Identify actionable targets and trial matches.

7. Output: Pan-cancer classification, driver report, pathway profiles, treatment recommendations.

Example Usage

User: "Classify this breast cancer patient's tumor in the pan-cancer context and identify shared drivers."

Agent Action:

python3 Skills/Oncology/Pan_Cancer_MultiOmics_Agent/pancancer_analyzer.py \
    --mutations patient_mutations.maf \
    --expression patient_rnaseq.tsv \
    --methylation patient_methylation.tsv \
    --cnv patient_cnv_segments.tsv \
    --reference tcga_pancancer \
    --subtype_method nmf_consensus \
    --output pancancer_report/

Pan-Cancer Molecular Subtypes

Cross-cancer molecular taxonomy identifies patterns beyond histology:

Subtype	Characteristics	Example Cancers
C1-Wound healing	High proliferation, MYC amp	Breast, ovarian, bladder
C2-IFN-gamma dominant	Immune active, high TCR/BCR	Melanoma, lung, cervical
C3-Inflammatory	NF-kB, cytokine signatures	Head/neck, stomach
C4-Lymphocyte depleted	Low immune, PTEN loss	Glioma, uveal melanoma
C5-Immunologically quiet	Low expression overall	Kidney chromophobe, thyroid
C6-TGF-beta dominant	High TGF-B, fibrosis	Pancreas, rectum, glioma

Deep Learning Architecture

Multi-Omics Integration Model:

Input Layers:
  - Genomic encoder (mutations, CNV)
  - Transcriptomic encoder (mRNA, miRNA)
  - Epigenomic encoder (methylation)
  - Proteomic encoder (RPPA)

Fusion Layer:
  - Cross-attention mechanism
  - Multi-modal variational autoencoder

Output Heads:
  - Subtype classifier
  - Survival predictor
  - Drug response predictor

MLOmics Database Access

The agent integrates with MLOmics, providing:

• 8,314 patient samples across 32 cancer types
• Pre-computed features for ML benchmarking
• Standardized train/test splits for reproducibility
• Drug sensitivity data for 300+ compounds

Prerequisites

• Python 3.10+
• PyTorch with multi-modal architectures
• Access to TCGA, CPTAC, or local data
• 16GB+ RAM for pan-cancer analysis

Related Skills

• Tumor_Clonal_Evolution - For intratumoral heterogeneity
• Multi_Omics_Integration - For single-patient integration
• Drug_Repurposing - For therapeutic matching

Clinical Applications

1. Cancer of Unknown Primary (CUP): Identify tissue of origin
2. Cross-indication trials: Find basket trial eligibility
3. Driver prioritization: Pan-cancer functional evidence
4. Prognosis: Multi-omics survival models

Author

AI Group - Biomedical AI Platform