PhuckCancer

PhuckCancer provides visualization, analysis, AI interpretation, evidence auditing, and human-reviewable reporting for large-scale cancer genomics, molecular evidence, mutation, pathway, treatment-response, and clinical-trial signal datasets.

It's really simple but powerful:

• Cancer data goes in.
• MAMMAL compares it against its learned biological patterns.
• MAMMAL outputs molecular interpretation.
• PhuckCancer checks the evidence.
• The local LLM explains it to humans.

Creator and author: Jose Perez. GitHub: dotdigitize. License: Apache License 2.0. Copyright 2026 Jose Perez.

PhuckCancer is the full cancer evidence platform. MAMMAL is the required biomedical AI engine inside it. The local LLM is the plain-English assistant that explains MAMMAL's structured interpretation. The evidence portal is the interface doctors, researchers, patients, and families use. Optional external data connectors allow PhuckCancer to retrieve and normalize cancer genomics data from supported research portals and APIs.

Core Mission

PhuckCancer exists to help doctors, researchers, patients, and families understand cancer evidence faster, organize it better, and fight cancer with stronger information.

The platform brings together cancer genomics data, molecular evidence, mutation and pathway analysis, a MAMMAL-powered biomedical AI engine, local LLM explanation, evidence auditing, optional external cancer genomics data retrieval, and human-reviewable reporting into one open-source cancer intelligence system.

PhuckCancer is not just a cancer data viewer and it is not just an AI chatbot. It is being built as a complete cancer research and evidence platform that can visualize cancer datasets, analyze molecular findings, route cancer evidence through MAMMAL-driven biomedical interpretation, explain complex medical language in plain English, flag unsupported claims, and produce reports that doctors, researchers, patients, and families can review together.

The goal is to make cancer information easier to see, easier to understand, easier to verify, and easier to act on with qualified medical professionals. Patients and families should not be lost inside confusing pathology reports, genomic test results, treatment notes, clinical-trial language, and scattered research data. Doctors and researchers should have better tools for reviewing cancer evidence, finding important signals, and explaining what matters.

Who It Helps

• Doctors and tumor boards reviewing cancer evidence, molecular findings, and audit trails.
• Researchers organizing cancer research, tumor genomics, molecular evidence, clinical trial evidence, drug resistance, and precision oncology signals.
• Patients and families preparing better questions and family-friendly summaries for qualified medical review.

Main Workflow

Cancer genomics data, report text, molecular evidence, external connector data, or MAMMAL output
-> parser
-> cancer finding extraction
-> MAMMAL biomedical interpretation layer
-> claim extraction
-> evidence matching
-> support scoring
-> risk flagging
-> local LLM doctor or family explanation
-> human review status
-> Markdown/JSON report export

Product Areas

PhuckCancer includes a cancer evidence portal, genomic alteration matrix, cancer pathway explorer, MAMMAL biomedical AI engine, local LLM patient and family assistant, evidence auditing, clinical-trial signal organizer, drug resistance signal watcher, optional cBioPortal data connector, and doctor/family report builder.

The MAMMAL research pipeline includes app/mammal_engine.py, app/mammal_pipeline.py, app/mammal_providers.py, app/mammal_api_client.py, app/mammal_importer.py, app/mammal_output_parser.py, and app/mammal_claim_extractor.py. Tests pass without live MAMMAL installed by mocking provider behavior; runtime biomedical interpretation still requires local MAMMAL or a configured MAMMAL API provider.

Why MAMMAL Is Required

MAMMAL is the required biomedical reasoning engine inside PhuckCancer. PhuckCancer uses MAMMAL to interpret cancer molecular evidence through its learned biological representation across proteins/biologics, small molecules/chemistry, and gene/transcriptomic data.

PhuckCancer applies MAMMAL in a cancer evidence workflow. Cancer genomics records, mutation data, pathway evidence, resistance signals, treatment-response notes, clinical-trial signals, and normalized external cancer data are routed into the MAMMAL-powered interpretation layer. MAMMAL produces structured biomedical interpretation that is then audited, risk-flagged, explained by the local LLM, and turned into human-reviewable reports.

PhuckCancer does not replace MAMMAL with fake or deterministic biomedical output. If MAMMAL is not installed locally and no MAMMAL API provider is configured, PhuckCancer fails closed and reports that the biomedical engine is unavailable.

The local LLM does not replace MAMMAL. The local LLM explains MAMMAL's structured biomedical interpretation in plain English for doctors, researchers, patients, and families.

MAMMAL Provider Modes

PhuckCancer can use MAMMAL in two provider modes:

1. Local MAMMAL provider

The local provider loads MAMMAL from the installed Python package and model files on the same machine.

2. MAMMAL API provider

The API provider connects to a MAMMAL service running on another local server, institutional server, private research machine, or dedicated biomedical model server. This allows PhuckCancer to use a larger MAMMAL deployment without installing the full model on the web application server.

Example API config:

MAMMAL_REQUIRED=true
MAMMAL_PROVIDER=api
MAMMAL_API_BASE_URL=http://localhost:9000
MAMMAL_API_TOKEN=
MAMMAL_API_INTERPRET_PATH=/v1/interpret
MAMMAL_API_HEALTH_PATH=/health

Role-Based Local LLM Explanations

PhuckCancer asks what type of user is using the system before generating local LLM explanations.

The same MAMMAL interpretation can be explained differently depending on the user:

• Patient or family member: plain English, appointment questions, uncertainty, and "ask your oncologist" language.
• Doctor or tumor board reviewer: molecular evidence, pathway context, risk flags, support scores, and clinical review questions.
• Cancer researcher: mechanisms, evidence gaps, research hypotheses, and reproducible notes.
• Data engineer or system administrator: data source status, MAMMAL provider status, MariaDB setup, cBioPortal connector configuration, and API behavior.

The local LLM does not create the biomedical interpretation. MAMMAL does. The local LLM formats and explains MAMMAL's structured output for the selected user type.

How MAMMAL and the Local LLM Work Together

MAMMAL does not replace the normal local LLM. MAMMAL is the biomedical AI engine that reads cancer-related biological and molecular data. The local LLM is the explanation layer that turns the structured biomedical interpretation into language that doctors, researchers, patients, and families can understand.

The system is designed around this flow:

Cancer data / molecular data
-> MAMMAL biomedical AI engine
-> structured biological interpretation
-> evidence audit and risk flagging
-> local LLM explanation
-> doctor report or family-friendly summary

MAMMAL is used to interpret biological meaning from data such as:

• Gene mutations such as TP53, KRAS, EGFR, BRAF, BRCA1, and BRCA2 alterations.
• Gene expression signals such as overexpression, underexpression, activation, or suppression.
• Pathway signals such as EGFR signaling, MAPK, PI3K/AKT, DNA repair, apoptosis, immune signaling, angiogenesis, and drug resistance pathways.
• Protein-related notes, molecular target information, and biological pathway evidence.
• Small molecule or drug-related evidence, including drug-target relationship notes when available.
• Cancer evidence notes from studies, reports, mutation records, pathway notes, resistance notes, and clinical-trial signal records.

MAMMAL should return structured biomedical interpretation, not a final treatment decision. Example structured output may look like this:

{
  "gene": "KRAS",
  "variant": "G12C",
  "cancer_type": "lung adenocarcinoma",
  "pathway": "MAPK signaling",
  "biological_signal": "KRAS G12C may activate downstream MAPK pathway signaling.",
  "evidence_strength": "moderate",
  "uncertainty": "Requires clinical and molecular review.",
  "possible_research_questions": [
    "Is this alteration linked to treatment resistance?",
    "Are there co-mutations affecting interpretation?",
    "Is there trial evidence involving this biomarker?"
  ],
  "risk_flags": [
    "Do not treat this as a treatment recommendation without oncologist review."
  ]
}

The local LLM then explains that structured interpretation in the right mode.

For a doctor or researcher, the local LLM may produce a concise medical-review summary:

KRAS G12C is a molecular alteration that may activate MAPK pathway signaling. The available evidence should be reviewed in the context of cancer type, co-mutations, stage, treatment history, prior resistance, available clinical guidelines, and clinical-trial evidence. This finding should be interpreted by the oncology team or molecular tumor board before it affects care decisions.

For a patient or family member, the local LLM may produce a plain-English explanation:

This report mentions a change in a gene called KRAS. KRAS helps control cell growth signals. Some KRAS changes can matter in cancer care, but this result does not automatically mean one specific treatment is right. A good question to ask the oncologist is: “Does this KRAS mutation affect treatment options, resistance, or clinical trials for this cancer type?”

In one sentence: MAMMAL interprets the biological meaning of cancer genes, mutations, proteins, molecules, expression signals, pathways, and drug/evidence relationships; then the local LLM explains that interpretation clearly for doctors, researchers, patients, and families.

How It Helps Doctors

For doctors, oncologists, molecular tumor boards, and researchers, PhuckCancer is designed to reduce the time it takes to review complex cancer evidence. The system can organize cancer genomics records, route molecular findings through the MAMMAL biomedical AI engine, extract claims, compare those claims against evidence, flag uncertainty, and generate doctor-reviewable reports.

A doctor-facing workflow may look like this:

Cancer genomic report or external dataset
-> mutation and pathway extraction
-> MAMMAL biomedical interpretation
-> evidence matching
-> risk and uncertainty flags
-> tumor board style report
-> qualified medical review

The doctor report can help highlight:

• Which genes, variants, biomarkers, or pathways appear important.
• Whether a molecular finding may be connected to treatment response, resistance, or clinical-trial evidence.
• Whether the evidence is supported, partially supported, unsupported, contradicted, or missing.
• Whether a claim uses unsafe language such as overconfident treatment language or diagnostic overreach.
• What information may be missing from the report.
• What should be reviewed by an oncologist, pathologist, genetic counselor, tumor board, or research coordinator.

The system should not tell a doctor what treatment to choose. It should help organize the evidence so a qualified medical team can make a better-informed decision.

How It Helps Patients and Families

For patients and families, PhuckCancer is designed to make cancer documents less confusing. A family member may not understand terms such as mutation, biomarker, pathway, amplification, deletion, drug resistance, immunotherapy, targeted therapy, clinical trial, or molecular tumor board. The local LLM assistant can explain those terms in plain English and help the family prepare better questions for the doctor.

A patient/family workflow may look like this:

Pathology report, genomic test, oncology note, or trial description
-> plain-English explanation
-> important genes and findings summary
-> missing information checklist
-> questions to ask the doctor
-> family-friendly summary for medical review

The assistant can help families ask questions such as:

• What type of cancer is this and what stage is it?
• Did the test find any mutations or biomarkers that matter?
• Are any of these findings connected to targeted therapy, immunotherapy, resistance, or clinical trials?
• Should we ask about a molecular tumor board review?
• Is more testing needed to confirm this finding?
• Are there co-mutations that change the meaning of this result?
• What does this report not tell us?
• What should we ask at the next oncology appointment?

The patient/family assistant should never say “take this treatment.” It should say “ask your oncologist about this finding” or “this may be worth discussing with your care team.”

Example: Family Question, Doctor Review, and Possible Care Change

This example is not a treatment recommendation. It shows how PhuckCancer could help a family ask better questions and help a doctor review evidence more quickly.

A family member may ask PhuckCancer:

My mother’s lung cancer report says KRAS G12C. What does this mean, and what should we ask the doctor?

PhuckCancer may explain:

KRAS is a gene involved in cell growth signaling. G12C is a specific KRAS alteration. Some KRAS alterations can matter in cancer care, but the meaning depends on the exact cancer type, stage, full molecular report, prior treatment history, co-mutations, and available clinical evidence. Ask the oncologist whether this KRAS G12C finding changes treatment options, clinical-trial options, or resistance review for this cancer type.

The family may then ask the oncologist:

Does this KRAS G12C mutation change treatment options, clinical-trial options, or resistance concerns for this cancer type? Should this case be reviewed by a molecular tumor board?

A doctor might answer:

This is an important molecular finding. We need to review it in the context of the full diagnosis, stage, prior treatments, co-mutations, pathology, performance status, and current clinical guidelines. We may order confirmatory testing, review the full genomic panel, check for other biomarkers, discuss this at a molecular tumor board, or consider whether targeted therapy, immunotherapy, chemotherapy, radiation, surgery, or a clinical trial is appropriate.

Depending on the cancer type, the full medical history, the full biomarker profile, and the doctor’s judgment, this kind of evidence review might lead to a care change such as:

• Ordering additional molecular testing.
• Confirming a biomarker or mutation with another test.
• Referring the case to a molecular tumor board.
• Reviewing targeted therapy options.
• Reviewing immunotherapy relevance.
• Reviewing clinical-trial options.
• Checking for drug resistance signals.
• Changing the treatment plan when the evidence supports a different approach.
• Monitoring response more closely.

In some real-world cancer cases, identifying an actionable biomarker and matching it with the right medically approved therapy or clinical-trial strategy can lead to a strong response, long-term disease control, or remission. PhuckCancer does not promise remission and does not decide treatment. Its goal is to help important evidence get noticed, explained, audited, and discussed with the qualified medical team that can make real treatment decisions.

Example: Doctor and Family Report Output

PhuckCancer can generate two different summaries from the same evidence.

Doctor/research report:

Finding: KRAS G12C alteration detected.
Biological context: KRAS signaling is associated with downstream MAPK pathway activation.
Evidence status: Needs qualified review with cancer type, stage, co-mutations, prior treatment history, and clinical evidence.
Risk flags: Do not treat as a standalone treatment recommendation.
Recommended review: Oncology team, molecular tumor board, and clinical-trial evidence review if appropriate.

Family-friendly summary:

The report mentions a change in a gene called KRAS. This may matter because KRAS is involved in cancer cell growth signals. This does not automatically mean one treatment is right, but it is important enough to ask the doctor about. Ask whether this finding affects treatment options, clinical-trial options, or whether the case should be reviewed by a molecular tumor board.

The same system can produce both outputs while keeping a clear safety boundary: the doctor reviews the evidence and makes medical decisions; the family gets help understanding what to ask.

Optional cBioPortal Data Connector

PhuckCancer can optionally connect to a public, institutional, or self-hosted cBioPortal API instance as an external cancer genomics data source.

This connector is used only for data retrieval and normalization. PhuckCancer remains its own cancer evidence platform, with its own MAMMAL-powered biomedical AI engine, local LLM assistant, evidence auditing workflow, risk flagging, and doctor/family report builder.

The default public API base URL is:

https://www.cbioportal.org/api

Self-hosted or institutional instances can be configured with:

ENABLE_CBIOPORTAL_CONNECTOR=true
CBIOPORTAL_BASE_URL=http://localhost:8080/api
CBIOPORTAL_AUTH_TOKEN=

When enabled, PhuckCancer can retrieve cancer studies, cancer types, sample lists, molecular profiles, mutation records, and available clinical data from a cBioPortal API instance, normalize the records into the PhuckCancer data model, and route the findings into the MAMMAL-powered biomedical interpretation and evidence audit pipeline.

This connector is optional. Local development, tests, sample fixtures, and non-interpretation workflows must work without cBioPortal, internet access, authentication, MariaDB, or Ollama. Biomedical interpretation workflows require MAMMAL.

Security and privacy requirements:

• Never commit API tokens.
• Store tokens only through environment variables.
• Do not log full tokens.
• Do not import real patient-identifiable data into sample fixtures.
• Follow institutional data access, data-use agreements, privacy rules, and local compliance.
• Treat external data as research/evidence input, not as a medical decision by itself.
• Any imported record must still go through MAMMAL-powered interpretation, evidence auditing, risk flagging, and qualified human review.

MAMMAL Install

MAMMAL stands for Molecular Aligned Multi-Modal Architecture and Language.

Official code: https://github.com/BiomedSciAI/biomed-multi-alignment

Hugging Face model: https://huggingface.co/ibm-research/biomed.omics.bl.sm.ma-ted-458m

Paper: https://arxiv.org/abs/2410.22367

pip install biomed-multi-alignment[examples]

Alternative source install:

git clone https://github.com/BiomedSciAI/biomed-multi-alignment.git
pip install -e ./biomed-multi-alignment[examples]

Documentation-only loading example:

from mammal.model import Mammal

model = Mammal.from_pretrained("ibm/biomed.omics.bl.sm.ma-ted-458m")
model.eval()

If that import path changes upstream, confirm the latest official usage from the MAMMAL repository. The application itself runs tests without live MAMMAL.

MariaDB Demo Database

PhuckCancer uses MariaDB/MySQL for its demo database and persistence layer. SQLite is not used.

sudo apt install mariadb-server mariadb-client
mysql -u root -p

CREATE DATABASE phuckcancer_demo CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
CREATE USER 'phuckcancer_user'@'localhost' IDENTIFIED BY 'change_this_password';
GRANT ALL PRIVILEGES ON phuckcancer_demo.* TO 'phuckcancer_user'@'localhost';
FLUSH PRIVILEGES;

mysql -u phuckcancer_user -p phuckcancer_demo < db/schema.sql
mysql -u phuckcancer_user -p phuckcancer_demo < db/seed_cancer_demo.sql

Medical Safety Notice

PhuckCancer is not a medical device, not a diagnostic system, and not a treatment recommendation engine.

It does not diagnose cancer, prescribe treatment, predict individual outcomes, determine clinical-trial eligibility, replace oncologists, replace molecular tumor boards, replace pathologists, replace genetic counselors, or replace licensed clinical judgment.

All outputs are for education, research support, evidence organization, and qualified human review.

Patients and families should use PhuckCancer to understand information and prepare better questions, not to make medical decisions without a qualified doctor.

Local Development Setup

Backend:

python3 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
python -m pytest
uvicorn app.main:app --host 0.0.0.0 --port 8717 --reload

Frontend:

npm install
npm run build
npm run dev

Running the Interface

Backend:

uvicorn app.main:app --host 0.0.0.0 --port 8717 --reload

Frontend:

npm run dev

Open:

http://SERVER-IP:5179

Backend API:

http://SERVER-IP:8717

Ollama optional:

ollama pull gemma4:e4b

MariaDB:

sudo apt install mariadb-server mariadb-client
mysql -u root -p

CREATE DATABASE phuckcancer_demo CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
CREATE USER 'phuckcancer_user'@'localhost' IDENTIFIED BY 'change_this_password';
GRANT ALL PRIVILEGES ON phuckcancer_demo.* TO 'phuckcancer_user'@'localhost';
FLUSH PRIVILEGES;

mysql -u phuckcancer_user -p phuckcancer_demo < db/schema.sql
mysql -u phuckcancer_user -p phuckcancer_demo < db/seed_cancer_demo.sql

API Overview

Core endpoints include /health, /api/system/status, /api/sample/overview, /api/data-sources, /api/cancer-records, /api/genomics/matrix, /api/pathways, /api/mammal/interpret, /api/evidence/audit, /api/assistant/explain, /api/reports/build, and optional /api/cbioportal/* endpoints.

Future Vision: AGI-Assisted Cancer Research System

The long-term goal of PhuckCancer is to become a self-expanding open-source cancer research and AGI-assisted evidence system that helps doctors, researchers, patients, and families fight cancer with better information, faster review, and stronger evidence organization.

PhuckCancer is being designed as more than a static cancer information portal. The future vision is to incorporate advanced AI and AGI-style systems that can continuously organize cancer research, compare molecular findings, review clinical evidence, analyze cancer genomics datasets, track treatment-resistance signals, explain complex reports, and help human experts identify important research directions faster.

The goal is not to replace doctors, oncologists, researchers, molecular tumor boards, pathologists, genetic counselors, or licensed medical judgment. The goal is to build an open-source cancer intelligence system that helps qualified humans see more, miss less, understand faster, and make better-informed decisions.

Future AGI-assisted capabilities may include:

• Self-expanding cancer research knowledge maps
• Automated review of new cancer genomics studies
• MAMMAL-powered biomedical interpretation workflows
• Local LLM assistants for doctors, researchers, patients, and families
• Molecular evidence comparison across cancer types
• Drug-resistance signal monitoring
• Clinical-trial evidence organization
• Plain-English explanations of cancer reports
• Question builders for oncology appointments
• Research hypothesis generation for human review
• Evidence grading and uncertainty tracking
• Open-source collaboration tools for cancer research communities
• Optional external cancer data connector expansion
• Multi-agent research review workflows
• Automated evidence-change tracking
• Human-reviewed cancer knowledge graph expansion

The mission is to build a complete open-source cancer research and AI evidence platform that helps people beat the odds against cancer by making cancer data easier to understand, easier to audit, easier to connect, and harder to ignore.

Project Structure

Backend code lives in app/, React code in src/, MariaDB scripts in db/, deterministic sample fixtures in sample_data/, and tests in tests/.

License

Apache License 2.0.

Attribution

See ATTRIBUTION.md.