misha

The misha package is a toolkit for analysis of genomic data. it implements an efficient data structure for storing genomic data, and provides a set of functions for data extraction, manipulation and analysis.

Installation

You can install the released version of misha from CRAN with:

install.packages("misha")

Or from conda:

conda install -c aviezerl r-misha

And the development version from GitHub with:

remotes::install_github("tanaylab/misha")

Usage

See the Genomes vignette for instructions on how to create a misha database for common genomes.

See the user manual for more usage details.

Using misha with an LLM agent

For agents (Claude Code, Copilot, Cursor, etc.) writing misha analysis code in a downstream project, point them at the maintained agent guides in this repo:

• agent-guides/misha-core.md
  • concepts, bootstrap, and the everyday recipes (intervals, annotation, distance, extract, vtracks, gscreen, gdist, gtrack.create). Start here.
• agent-guides/misha-advanced.md
  • 2D / Hi-C pile-ups, insulation, sequence and PWM tracks, bulk import/export, new genomes and cross-species.
• agent-guides/misha-anti-patterns.md
  • silent footguns referenced inline from the above.
• agent-guides/skills/
  • deep playbooks for specific tasks. Currently: importing-tracks (format chooser across all gtrack.*import* variants + pre/post-import validation). Load when the task specifically calls for one of these.

The core guide is ~4k words and targets a system-prompt-sized context. For Claude Code-style setups, dropping misha-core.md (or all three) into the project’s CLAUDE.md / AGENTS.md is the intended use.

Drop-in prompt (no clone needed). Paste the block below into your agent at the start of a misha task. It points the agent at the raw files on GitHub, so it works without a local checkout:

Before writing any misha code, fetch and read:

- https://raw.githubusercontent.com/tanaylab/misha/master/agent-guides/misha-core.md  (mandatory: concepts + everyday recipes)
- https://raw.githubusercontent.com/tanaylab/misha/master/agent-guides/misha-anti-patterns.md  (silent footguns; cross-referenced from core)
- https://raw.githubusercontent.com/tanaylab/misha/master/agent-guides/misha-advanced.md  (consult on demand: 2D / Hi-C, PWM, import/export, new genomes)

Follow the conventions in those files. When you hit a recipe with an
"Avoid:" block, treat it as a hard rule.

Pin to a release tag for stability by replacing master with any tag that contains agent-guides/ (the directory landed in v5.7.0, so v5.7.0 or later; v5.7.3 for the current guide content).

Running scripts from old versions of misha (< 4.2.0)

Starting in misha 4.2.0, the package no longer stores global variables such as ALLGENOME or GROOT. Instead, these variables are stored in a special environment called .misha. This means that scripts written for older versions of misha will no longer work. To run such scripts, either add a prefix of .misha$ to all those variables (.misha$ALLGENOME instead of ALLGENOME), or run the following command before running the script:

ALLGENOME <<- .misha$ALLGENOME
GROOT <<- .misha$GROOT
ALLGENOME <<- .misha$ALLGENOME
GINTERVID <<- .misha$GINTERVID
GITERATOR.INTERVALS <<- .misha$GITERATOR.INTERVALS
GROOT <<- .misha$GROOT
GWD <<- .misha$GWD
GTRACKS <<- .misha$GTRACKS