15% of all profit is donated to Heal Palestine and the Palestine Children's Relief Fund (PCRF)

Computational oncology · one RNA-seq profile

Cancer hides in the transcriptome.

Provotics reads a tumor's RNA-seq profile and returns its full molecular portrait: where in the body it arose, its molecular subtype, its candidate druggable alterations, and its immune profile, with the genes behind every call.

25anatomical sites
17,410tumor profiles
~18kgenes per profile
~94%balanced accuracy

Trained and benchmarked on 17,410 public tumor RNA-seq profiles spanning 25 anatomical sites.

The problem

A tumor doesn't always announce where it came from

Cancers are defined by where they start, but metastases and cancers of unknown primary can hide that origin, and the answer changes how a patient is treated. The signal is written in the tumor's gene expression. Reading it is the whole game.

Origin drives treatment

Where a cancer began shapes the entire treatment plan, yet metastatic and unknown-primary tumors blur that answer.

High-dimensional data

Each sample carries tens of thousands of gene measurements. The biological signal is real but buried in noise and correlation.

Interpretability is not optional

In oncology, a prediction is only actionable with its reasoning and honest uncertainty attached. The genes behind a call are what make it reviewable.

What it does

Diagnose. Trust. Treat.

Provotics turns one expression profile into a calibrated, explainable call, and points toward what to do next.

Diagnose

Reads a tumor's site of origin across 25 anatomical sites straight from its transcriptome, ~94% balanced accuracy on held-out tumors, then reads its molecular subtype, likely alterations, and immune profile.

Trust

Every call carries a calibrated probability, so a stated 80% means roughly 80% in practice. Low-confidence or out-of-distribution samples are abstained on or flagged as novel, rather than forced into a label.

Treat

Surfaces the tumor's over-expressed, druggable targets as research hypotheses, a bridge from a diagnosis toward candidate therapies.

How it works

From raw counts to a readable portrait

One pipeline, four steps. Each profile is quality-checked, harmonized to a common gene space, scored by the model, and returned with its reasoning attached.

RNA-seq in

Start from a single tumor expression profile, gene-level counts or TPM.

QC & harmonize

Mapped to a common gene space, normalized, and screened for quality before any prediction.

Model & calibrate

An ensemble scores the profile and calibrates the probabilities, abstaining when confidence is low.

Portrait out

Site of origin, subtype, druggable targets, and immune profile, with the driver genes behind each call.

Live sample report

See what one profile turns into

Pick an illustrative tumor profile and watch the portrait Provotics produces, the calibrated site call, its top alternatives, the molecular detail, and the driver genes behind it.

Choose a sample

Predicted site of origin

Prostate 93% confidence

Calibrated probability. Top alternatives shown below.

Molecular detail

Molecular subtypen/a
Immune profilen/a
Candidate targets
Driver genes

Illustrative examples for demonstration. Sample profiles, probabilities, and markers are representative, not live model outputs, and no proprietary data or model weights are shipped to your browser. The real model runs server-side for approved users.

Why trust it

Validated, calibrated, and honest about its limits

Numbers from held-out tumors and independent cohorts, not the training set. Where the model is weak, it says so rather than guessing.

92.6%accuracy on 1,724 held-out tumors
0.96site accuracy on an independent cohort (n=381)
0.14 → 0.03calibration error, after temperature scaling
0.87macro-F1 across 25 anatomical sites
Fully independent, out-of-distribution cohorts show the generalization gap that honest external validation always reveals. Rather than force a label, Provotics runs a novelty detector and abstains on samples it doesn't recognize, so a wrong call is caught instead of reported. Provotics is a research and educational project, not a medical device.
Who it's for

Built for the people reading tumors

From a single RNA-seq profile to a portrait you can act on, wherever expression data is generated and interpreted.

Research labs

Characterize tumor samples and unknown-primary cases, and generate testable hypotheses straight from expression data.

Biotech & drug discovery

Surface over-expressed, druggable targets across cohorts to prioritize programs and narrow the search.

Computational oncology

A calibrated, explainable baseline with the driver genes exposed, so every call is reviewable, not a black box.

FAQ

Questions, answered

Is Provotics a medical device or diagnostic tool?
No. Provotics is a research and educational project. It is not a medical device and is not intended for clinical diagnosis or treatment decisions.
What data was it trained on?
17,410 public tumor RNA-seq profiles spanning 25 anatomical sites, with roughly 18,000 genes per profile. Benchmarks are reported on held-out tumors and independent cohorts the model never saw during training.
How accurate is it, really?
92.6% accuracy and 0.87 macro-F1 on held-out tumors, and 0.96 site accuracy on an independent cohort. Fully out-of-distribution cohorts show a larger gap, which is expected, and the model abstains on samples it doesn't recognize rather than guessing.
What happens when it isn't sure?
Every call carries a calibrated probability. Low-confidence or out-of-distribution profiles are flagged as novel or abstained on, instead of being forced into a label. Try "Sample D" in the demo to see it in action.
How do I get access?
The model is invite-only while we finalize it. Join the waitlist below and we'll reach out when your access is ready. It's free for academic research, with commercial tiers coming later.
KB

Kareem Badran

Biology + AI @ UNC Charlotte

Provotics is my deep dive into computational oncology, bridging wet-lab biology and computation to read cancer's molecular signature. Open to research-lab and biotech collaborations.

Request access

Apply for access to Provotics

Access is invite-only and granted under a confidentiality agreement. Tell us who you are and how you plan to use the model, and review the access terms before submitting.

Free for academic research Commercial tiers coming soon