Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Tumor Heterogeneity Agent · FreedomIntelligence | DecimalAI

Tumor Heterogeneity Agent

SkillScore73Maintained

tumor-heterogeneity-agent

AI-powered intratumor heterogeneity analysis for clonal architecture reconstruction, subclonal evolution tracking, and therapy resistance prediction using multi-region and longitudinal sequencing.

scienceby FreedomIntelligence

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Clinical / Healthcare pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/tumor-heterogeneity-agent/SKILL.md

decimalai skills pull tumor-heterogeneity-agent --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/tumor-heterogeneity-agent/SKILL.md https://app.decimal.ai/s/tumor-heterogeneity-agent/SKILL.md

Raw file: /s/tumor-heterogeneity-agent/SKILL.md ↗ · machine-readable summary at /s/tumor-heterogeneity-agent.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 52d ago

Yes — +18% lift · at higher cost+50% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-07	✗→✓	▲ Improved	17%	0%
case-15	✗→✓	▲ Improved	86%	0%
case-16	✗→✓	▲ Improved	88%	0%
case-22	✗→✓	▲ Improved	38%	0%
case-08	✓→✓	＝ Same ✓	23%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/1/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

FreedomIntelligence / OpenClaw-Medical-Skills

skills/tumor-heterogeneity-agent/SKILL.md

View on GitHub ↗

Imported 8/1/2026Version v2

Skill contents

SKILL.md

<!--

COPYRIGHT NOTICE

This file is part of the "Universal Biomedical Skills" project.

Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>

All Rights Reserved.

This code is proprietary and confidential.

Unauthorized copying of this file, via any medium is strictly prohibited.

Provenance: Authenticated by MD BABU MIA

-->

Tumor Heterogeneity Agent

The Tumor Heterogeneity Agent provides comprehensive analysis of intratumor heterogeneity (ITH) for understanding clonal architecture, tracking subclonal evolution, and predicting therapy resistance. It integrates multi-region sequencing, single-cell data, and longitudinal samples to reconstruct tumor phylogenies and identify actionable subclones.

When to Use This Skill

When analyzing multi-region tumor sequencing for clonal architecture.
For tracking clonal evolution under treatment pressure.
To predict resistance emergence from subclonal populations.
When assessing tumor heterogeneity impact on treatment response.
For integrating single-cell and bulk sequencing for ITH analysis.

Core Capabilities

Clonal Deconvolution: Infer clonal populations and their frequencies.

Phylogeny Reconstruction: Build tumor evolutionary trees from variants.

Subclonal Tracking: Monitor subclone dynamics over time.

Resistance Prediction: Identify pre-existing resistant subclones.

Multi-Region Integration: Combine spatial heterogeneity data.

Single-Cell ITH: Integrate scDNA-seq for ground-truth clones.

Heterogeneity Metrics

| Metric | Definition | Clinical Relevance | |--------|------------|-------------------| | MATH Score | Mutant-allele tumor heterogeneity | ITH quantification | | Shannon Index | Clonal diversity | Evolutionary potential | | Clone Count | Number of distinct clones | Complexity | | Truncal Fraction | % truncal mutations | Targetability | | ITH Score | Composite heterogeneity | Prognosis |

Workflow

Input: Multi-region/longitudinal WES/WGS, copy number, tumor purity.

Preprocessing: Variant calling, CNV calling, purity estimation.

CCF Estimation: Calculate cancer cell fraction for each mutation.

Clustering: Group mutations into clonal populations.

Phylogeny: Reconstruct evolutionary tree.

Temporal Analysis: Track clone dynamics over time.

Output: Clone structures, phylogenies, heterogeneity metrics.

Example Usage

User: "Analyze the clonal architecture of this multi-region lung tumor sequencing to understand heterogeneity and identify resistant subclones."

Agent Action:

bash
python3 Skills/Oncology/Tumor_Heterogeneity_Agent/ith_analysis.py \
    --multi_region_vcfs region1.vcf,region2.vcf,region3.vcf \
    --cnv_segments cnv_calls.seg \
    --purity 0.7,0.65,0.72 \
    --sample_names Primary,Met1,Met2 \
    --method pyclone-vi \
    --phylogeny_method citup \
    --output ith_analysis/

Deconvolution Methods

| Method | Approach | Best For | |--------|----------|----------| | PyClone-VI | Variational inference | Large datasets | | SciClone | Kernel density | High purity | | EXPANDS | Probabilistic | Multi-region | | Canopy | EM algorithm | CNV integration | | Clonevol | Phylogeny-aware | Longitudinal | | CITUP | Integer programming | Tree optimization |

Input Requirements

| Input | Format | Required | |-------|--------|----------| | Somatic Variants | VCF with depth | Yes | | Copy Number | SEG file | Yes | | Tumor Purity | Float (0-1) | Yes | | Sample Metadata | TSV | Yes | | Normal BAM | BAM | Recommended |

Output Components

| Output | Description | Format | |--------|-------------|--------| | Clone Assignments | Mutation-to-clone mapping | .csv | | Clone Frequencies | Per-sample clone fractions | .csv | | Phylogenetic Tree | Newick and visualization | .nwk, .pdf | | ITH Metrics | Heterogeneity scores | .json | | Subclone Variants | Clone-specific mutations | .vcf | | Evolution Plot | Clone dynamics over time | .png | | Actionable Subclones | Druggable clone mutations | .csv |

Clonal Classification

| Clone Type | Definition | Implications | |------------|------------|--------------| | Truncal | Present in all samples | Ideal targets | | Branch | Present in subset | Regional targets | | Private | Single sample only | Local significance | | Resistant | Expand under therapy | Resistance mechanism |

AI/ML Components

Clone Inference:

Variational autoencoders for CCF estimation
Dirichlet process mixture models
Graph neural networks for phylogeny

Resistance Prediction:

Time-series models for clone trajectories
Classification of resistant signatures
Drug-clone interaction prediction

Multi-Region Integration:

Multi-task learning across regions
Spatial models for regional patterns
Transfer learning across cancers

Clinical Applications

| Application | ITH Insight | Clinical Action | |-------------|-------------|-----------------| | Treatment Selection | Truncal vs branch targets | Prioritize truncal targets | | Resistance Monitoring | Pre-existing resistant clones | Early combination therapy | | Prognosis | ITH score | Risk stratification | | Biomarker Development | Clonal biomarkers | Robust biomarker selection |

Cancer-Specific Patterns

| Cancer Type | Typical ITH | Key Drivers | |-------------|-------------|-------------| | Lung (NSCLC) | High | EGFR, KRAS subclonal | | Breast | Moderate-High | PIK3CA, ESR1 evolution | | Colorectal | Moderate | KRAS, BRAF clonal | | Renal | Very High | VHL truncal, diverse branches | | Melanoma | High | BRAF/NRAS truncal |

Prerequisites

Python 3.10+
PyClone-VI, SciClone
CITUP, Clonevol
CNVkit/FACETS for CNV
R with clonal evolution packages

Related Skills

ctDNA_Dynamics_MRD_Agent - Liquid biopsy tracking
Single_Cell_CNV_Agent - scDNA-seq analysis
HRD_Analysis_Agent - Genomic instability
Pan_Cancer_MultiOmics_Agent - Multi-omic integration

Phylogeny Visualization

| View Type | Shows | Best For | |-----------|-------|----------| | Fish Plot | Clone dynamics over time | Longitudinal | | Tree Diagram | Branching evolution | Multi-region | | Muller Plot | Population dynamics | Treatment response | | Clone Map | Spatial distribution | Multi-region spatial |

Special Considerations

Sampling Bias: Multi-region captures more heterogeneity
Purity Effects: Low purity reduces clone resolution
CNV Complexity: High CNV burden complicates CCF
Single-Cell Validation: Ground truth from scDNA-seq
Temporal Resolution: Frequent sampling improves tracking

Resistance Mechanisms

| Mechanism | Detection | Intervention | |-----------|-----------|--------------| | Pre-existing resistant clone | Subclonal at baseline | Combination therapy | | Acquired resistance | New clone emerges | Switch therapy | | Phenotypic plasticity | Expression change | Monitor phenotype | | Microenvironment | TME evolution | Immunotherapy |

Author

AI Group - Biomedical AI Platform

Illustrative Want to see what tumor-heterogeneity-agent would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/2/2026, 18:23 UTC · 22 cases

Lift

+18%

73% vs 55%

Agent turns

1.0 vs 1.0

Tokens

+50%

3,794.59 vs 2,537.36

fail → pass · 4pass → pass · 12fail → fail · 6

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-06 We are tracking longitudinal biopsies collected from an EGFR-mutant lung cancer patient before and after osimertinib resistance. Which deconvolution tool is specifically designed to be phylogeny-aware across temporal samples?	fail→fail	20,724	6,715	-68%	1	1	0%	3,430	2,876	-16%	0	0	—
case-07 Our tumor samples exhibit extensive copy-number alterations across multiple chromosome arms. We need an Expectation-Maximization framework to jointly model local copy number and mutation CCFs. Which deconvolution tool should we select?	fail→pass	14,861	7,259	-51%	1	1	0%	2,491	2,918	+17%	0	0	—
case-08 We have calculated mutation CCFs across 4 spatial regions and need to optimize the binary and non-binary phylogenetic tree topology using integer programming. Which specialized phylogeny reconstruction method should we specify in our pipeline?	pass→pass	12,840	4,421	-66%	1	1	0%	2,010	2,464	+23%	0	0	—
case-09 We want to compute a simple single-sample quantitative metric of mutant allele frequency distribution to summarize intratumor heterogeneity without full Bayesian clustering. Should we calculate the Gini coefficient or MATH score, and what does it measure?	pass→pass	11,809	10,825	-8%	1	1	0%	2,160	3,504	+62%	0	0	—
case-05 Our lab has high-purity (purity > 0.85) flow-sorted tumor samples from a single region. We want a fast non-parametric density method to cluster Cancer Cell Fractions without heavy integer linear programming tree constraints. Which deconvolution algorithm fits this high-purity single-region scenario best?	pass→pass	18,193	8,747	-52%	1	1	0%	3,023	3,254	+8%	0	0	—
case-14 We need to create a publication figure showing how four distinct tumor subclones expand and shrink over 12 months across 4 treatment cycles. Should we use a static radial phylogenetic tree or a Fish plot?	pass→pass	11,309	9,290	-18%	1	1	0%	1,771	3,350	+89%	0	0	—
case-15 To present the spatial arrangement of identified subclones across 8 grid-based biopsies of a localized liver tumor resection, which visualization format is most appropriate?	fail→pass	8,615	6,178	-28%	1	1	0%	1,439	2,677	+86%	0	0	—
case-01 I have longitudinal whole-exome sequencing data across three temporal biopsies (baseline, post-treatment, and relapse) from a breast cancer patient, along with copy number segment files and estimated tumor purities. Could you run an intratumor heterogeneity analysis to track how subclones evolve under treatment? I need the output to include CSV files detailing mutation-to-clone assignments and clone frequencies across samples, a Newick format phylogenetic tree, and an evolution plot showing subclone dynamics over time.	fail→fail	21,482	40,081	+87%	1	1	0%	4,116	6,947	+69%	0	0	—
case-02 We performed multi-region sequencing on a primary tumor and two metastatic lesions (VCFs: primary.vcf, met1.vcf, met2.vcf) accompanied by copy number calls (cnv_calls.seg) and purity estimates. Please analyze the spatial heterogeneity to distinguish truncal from branch mutations. I'm expecting a JSON file with heterogeneity metrics like MATH score and Shannon index, a CSV table listing actionable subclonal variants, and a PDF diagram showing the phylogenetic branching tree.	fail→fail	27,297	27,659	+1%	1	1	0%	5,615	7,957	+42%	0	0	—
case-03 I need to reconstruct the clonal architecture from multi-region sequencing of a lung adenocarcinoma specimen. I'll provide the VCF files for each region, the copy number segment file, and sample purity floats. Please deconvolve the clonal populations and provide CSV files mapping mutations to inferred clones, a frequency table per region, and a visual plot of the tree structure along with identified resistant or private subclones.	fail→fail	28,136	17,085	-39%	1	1	0%	6,221	4,836	-22%	0	0	—
case-04 We have multi-region sequencing covering 50 tumor regions with over 100,000 somatic variants. We want to perform fast clonal deconvolution without running into computational bottlenecks. Should we use Kernel Density estimation or a variational inference approach for clustering cancer cell fractions?	pass→pass	15,790	11,510	-27%	1	1	0%	2,556	3,567	+40%	0	0	—
case-10 When evaluating clonal diversity to gauge the overall evolutionary potential of a tumor, which ecological diversity metric should be computed from the clone proportions?	pass→pass	10,271	3,759	-63%	1	1	0%	1,757	2,250	+28%	0	0	—
case-11 We identified two actionable mutations in a metastatic melanoma specimen: BRAF V600E present in 100% of cancer cells across all biopsy regions, and MEK1 C121S present in only 15% of cells in a single biopsy. Which mutation is classified as truncal and how does this affect targeted therapy selection?	pass→pass	10,348	11,005	+6%	1	1	0%	1,894	3,767	+99%	0	0	—
case-12 In clear cell renal cell carcinoma (ccRCC), what is the typical overall level of intratumor heterogeneity across multi-region biopsies, and which gene is characteristically truncal versus subclonal?	pass→pass	8,727	8,092	-7%	1	1	0%	1,602	3,059	+91%	0	0	—
case-13 When analyzing multi-region sequencing of non-small cell lung cancer (NSCLC), what pattern of clonal architecture is typical regarding driver mutations like EGFR or KRAS?	fail→fail	10,973	11,097	+1%	1	1	0%	1,745	3,586	+106%	0	0	—
case-16 Baseline sequencing of a colorectal cancer patient revealed a tiny subclone (2% CCF) carrying KRAS G12D before anti-EGFR therapy. After 6 months of cetuximab, this subclone expanded to 80% CCF. How is this resistance mechanism classified, and how could it have been mitigated?	fail→pass	11,000	11,162	+1%	1	1	0%	1,887	3,540	+88%	0	0	—
case-17 When setting up input files for clonal CCF estimation using multi-region WES, is a somatic VCF alone sufficient, or are tumor purity estimates and copy number segmentation files required?	pass→pass	13,138	5,694	-57%	1	1	0%	2,383	2,713	+14%	0	0	—
case-18 A researcher claims that a single core needle biopsy is sufficient to capture the full spectrum of subclonal mutations in a 5cm heterogeneous tumor. What major analytical limitation invalidates this claim?	pass→pass	6,609	6,473	-2%	1	1	0%	1,081	2,761	+155%	0	0	—
case-19 We collected cell-free DNA (cfDNA) from plasma of a stage II colon cancer patient post-resection to detect circulating tumor DNA for minimal residual disease (MRD) monitoring using ultra-deep targeted sequencing with UMIs. Can you provide a tumor heterogeneity subclone reconstruction for this plasma cfDNA sample?	fail→fail	12,591	29,495	+134%	1	1	0%	2,206	7,609	+245%	0	0	—
case-20 A patient's family history shows multiple cases of early-onset breast and ovarian cancer. We sequenced peripheral blood germline DNA and found a BRCA1 frameshift variant. Can you apply the tumor heterogeneity deconvolution agent to classify this germline variant according to ACMG/AMP pathogenicity criteria?	pass→pass	10,955	9,565	-13%	1	1	0%	1,877	3,431	+83%	0	0	—
case-21 We need to calculate the composite Homologous Recombination Deficiency (HRD) score—combining Loss of Heterozygosity (LOH), Telomeric Allelic Imbalance (TAI), and Large-Scale State Transitions (LST)—from genome-wide SNP array data of an ovarian carcinoma. Does the tumor heterogeneity agent perform HRD scar scoring?	pass→pass	12,365	8,709	-30%	1	1	0%	2,284	3,286	+44%	0	0	—
case-22 When training machine learning models for CCF estimation in high-dimensional multi-region datasets, what neural network architecture can be used according to standard ITH AI workflows?	fail→pass	13,266	8,377	-37%	1	1	0%	2,274	3,129	+38%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 54.6%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v2
Run completed: 8/2/2026, 18:23 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Prompt

Our tumor samples exhibit extensive copy-number alterations across multiple chromosome arms. We need an Expectation-Maximization framework to jointly model local copy number and mutation CCFs. Which deconvolution tool should we select?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run73

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+53% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 51 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	7/26/2026	+23%

Other measured skills in the registry, with their headline benchmark lift.

Ready to supercharge your AI agents?

Tumor Heterogeneity Agent

SkillSafety

Source

COPYRIGHT NOTICE

This file is part of the "Universal Biomedical Skills" project.

Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>

All Rights Reserved.

This code is proprietary and confidential.

Unauthorized copying of this file, via any medium is strictly prohibited.

Provenance: Authenticated by MD BABU MIA

Tumor Heterogeneity Agent

When to Use This Skill

Core Capabilities

Heterogeneity Metrics

Workflow

Example Usage

Deconvolution Methods

Input Requirements

Output Components

Clonal Classification

AI/ML Components

Clinical Applications

Cancer-Specific Patterns

Prerequisites

Related Skills

Phylogeny Visualization

Special Considerations

Resistance Mechanisms

Author

How this skill was measured

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Related skills