FISH image analysis for oncogene amplification typing (ecDNA vs HSR).
AMPLIFISH turns multi-channel FISH microscopy images into per-nucleus feature tables — target/centromere spot counts, nucleus shape, and the Shannon entropy of the target-spot spatial distribution — that separate ecDNA (extrachromosomal DNA) from HSR (homogeneously staining regions). The pipeline is unsupervised: no labelled training data or fine-tuning required.
Full documentation lives in docs/, including a guide to using AMPLIFISH on your own data.
git clone https://github.com/akdemirlab/AMPLIFISH.git
cd AMPLIFISH
conda create -n amplifish python=3.9 -y && conda activate amplifish
pip install -e .Dependencies (including the CellposeSAM segmentation backend) install automatically. See docs/installation.md for system requirements and details.
cd demo && bash demo.sh # or: python run_demo.py --cpuRuns the full pipeline on two bundled Colo320 FISH images (1 ecDNA + 1 HSR) and writes per-nucleus feature tables plus a figure separating the two amplification types. See docs/demo.md for expected output and run time.
Pointers to where each required item lives:
- Source code: src/amplifish/
- Demo dataset: demo/data/raw/ — 2 real Colo320 FISH images (1 ecDNA + 1 HSR)
- System requirements (dependencies + OS + version numbers, versions tested, non-standard hardware): docs/installation.md
- Installation guide (instructions + typical install time): docs/installation.md
- Demo (instructions to run, expected output, expected run time): docs/demo.md, with reference results in demo/expected_output/
- Instructions for use (run on your own data): docs/usage.md
MIT License, Copyright (c) Akdemir Lab. See LICENSE.
If you use AMPLIFISH in your research, please cite the accompanying manuscript (citation to be added on publication).