Skip to content

Releases

Pytheas 1.1.0

Choose a tag to compare

@jwilliamson16 jwilliamson16 released this 06 Nov 22:58

The published Pytheas (D'Ascenzo et al., Nature Communications 13, 2424 (2022)),
packaged for current Python, with a new desktop GUI.

Install

You need Python 3.10 or newer.

python3 -m venv ~/venvs/pytheas1
source ~/venvs/pytheas1/bin/activate          # Windows: ~\venvs\pytheas1\Scripts\activate
pip install pytheas1

You can also install the wheel attached below: pip install pytheas1-1.1.0-py3-none-any.whl.

Try it

  1. Download pytheas1_demo.zip (below) and unzip it. It holds the paper's Training_set
    inputs and the original README's demo settings.
  2. Run pytheas1 --workdir pytheas1_demo and click Run all.
  3. All six stages take a few minutes. The final report lists the 137 Training_set
    identifications.

What changed from ldascenzo/pytheas

  • The analysis code is unchanged. It is the published library code, changed only where
    current Python, numpy, pandas, matplotlib and Biopython no longer run it: 30 lines added and
    19 removed in 7 files, each one listed in PORTING.md.
    • No algorithm, scoring, default or bug fix was changed.
    • Each stage was checked against the original code on its original stack. See
      VALIDATION.md.
  • New:
    • a pip package;
    • a run API and command line;
    • a PyQt5 GUI with the original forms' options and help text.
  • Sp is a median 3.5% lower than the values printed in the paper. This comes from the
    author's own 2022 fix to consecutive-series counting (6d299ca), which is already in the
    upstream code. The identifications are the same.

Known issues

  • A shared letter in the default nucleotide sheets. m1acp3Y and t6A share the letter
    o, so a modification file that places m1acp3Y gets t6A's mass. KNOWN_ISSUES.md gives the
    workaround.
  • Tested on macOS only. Windows and Linux are not tested yet; please report problems as
    issues.