pai-soma-data 1.0.3

A class to handle data from a Soma experiment made by Phenomic AI

Pai Soma Data

Description

This package provides a class, SomaData, for handling data from a Soma experiment produced by Phenomic using TileDBSOMA. It allows users to interact with cached observation data as a Pandas DataFrame and retrieve subsets of data as AnnData objects, with features to query raw or normalized counts.

Features

• Load and cache observation (obs) and variable (var) data from a SOMA experiment.
• Retrieve subsets of data based on specified rows and columns.
• Automatically chooses the appropriate layer (raw or norm) based on the number of genes requested.
• Converts queried data into AnnData format for compatibility with downstream analysis tools.

Installation

To use this package, you'll need Python 3.8 or later.

pip install pai-soma-data

Usage

Initializing the SomaData Class

To start using the SomaData class, initialize it with the URI of your SOMA experiment. You can choose the data layer (raw or norm) based on your needs:

from pai_soma_data import SomaData

# Instantiate SomaData with normalized counts
sdata = SomaData(
    corpus_uri="...",
    layer="norm"  # Change to "raw" for raw counts
)

Accessing Metadata

The observation and gene metadata are stored in the .obs and .var attributes of the SomaData object, respectively. These are accessible as Pandas DataFrames:

# View the first few rows of observation metadata
print(sdata.obs.head())

# View the first few rows of variable (gene) metadata
print(sdata.var.head())

Slicing X Data

The SomaData class allows for Pandas-style slicing of data using the format (row_filter, col_filter). The data is fetched in the layer specified during instantiation (default is "norm" for normalized data).

• Row filter: Can be a pd.Series, list of integers, or slices.
• Column filter: Can be a gene name (string) or a list of gene names.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request for any improvements or bug fixes.

License

This project is licensed under the MIT License.