DNA Entropy Graph

I am building a Windows app for lab biologists who want a per-base Shannon entropy track without opening a terminal. Researchers in a lab at Vanderbilt University Medical Center already use the command-line prototypes this succeeds; their PI greenlit this app to replace them. You drop in a GenBank or FASTA file, press Run, and get a number from 0 to 2 bits at every position showing how predictable that base looks to the Evo 2 genomic language model, ready to open in IGV, Geneious, SnapGene or Benchling.

How it works

• Evo 2 needs a GPU, so rather than asking you to own one or asking you to trust me with your sequences, the app rents a GPU machine for a few minutes inside your own Google Cloud project, runs the job there, downloads the results and cleans the machine up. There is no server of mine in the path and no account of mine on your project; the bill is yours and visible to you. docs/threat_model.md is the honest account of what that design does and does not expose.
• The science is a Python package in worker/ that ships as a container image. It reads the sequence, runs it through the model over a sliding window in both directions, turns the per-base probabilities into Shannon entropy, and writes bedGraph, wig, GFF3 and TSV tracks.
• The Windows app in app/ is C# and WinUI 3 and owns everything else: input validation, the job state machine, retry-safe cloud calls, cost tracking and a SQLite run history. It talks to Google Cloud through its own gateway layer, with an in-memory fake behind the same interfaces for tests.
• A run is usually well under a dollar. The cost people forget is that a stopped machine's disk keeps charging, so the app tracks every resource it created and offers to delete it.

Where it stands

Not shipped. There is no installer and no v1. Measured on 2026-09-20, Windows 11:

part	state	evidence
worker pipeline, stand-in predictor	runs end to end, no GPU needed	810 passed, 2 skipped
app shell, run history, run progress	builds and launches	327 tests, 0 failing, 1 skipped
Evo 2 on a real GPU	never run	no GPU on the machine it is written on
any call to a real Google Cloud project	never run	every cloud path is covered against a fake only

docs/ToTest.md is the queue of what has to be proven on real infrastructure before anyone trusts it there, and the issues are the only tracker.

Build and run

git clone https://github.com/Raaif-Yousuf/DNA-Entropy-Graph
cd DNA-Entropy-Graph\app && dotnet build -c Release -p:Platform=x64
py -m venv ..\worker\.venv && ..\worker\.venv\Scripts\pip install -e ..\worker[dev]
..\worker\.venv\Scripts\dna-entropy run --input ..\worker\tests\data\sample.gb --name demo --out out

Tests: dotnet test -c Release -p:Platform=x64 from app/, and worker\.venv\Scripts\python -m pytest worker/tests -m "not gpu" from the repo root.

To use the app rather than build it, start at docs/user_guide/README.md. To work on it, start at docs/onboarding.md and docs/README.md.

This succeeds two command-line prototypes I have since retired; the Python science package is ported from them, and the app and the direct-to-Google-Cloud design are new. FEATURES.md has the prototypes' own feature inventory. MIT licensed, see LICENSE.