LISA: Landscape In-Silico deletion Analysis

Table of Contents

About

LISA is a statistical test for the influence of Transcription Factors on a set of genes. We leverage integrative modeling of public chromatin accessiblity and factor binding to make predictions that go beyond simple co-expression analysis. The minimum you need to run LISA is a list of genes-of-interest, but you can also supply your own epigenetic background. For more information, see Qin et al., 2020. This implementation extends the original, running faster, reducing dependencies, and adding useful CLI functions for pipeline integration.

The python package is easy to install and has a rich set of features and options. For a quick introduction to the method, check out the web interface.

The Model -----

The key components of the LISA test are the:

profile, a distribution of accessibility over regions in the genome, supplied by user or predicted from public data
hits, the regions where a TF is predicted to bind (through ChIP-seq or motif)
region-gene map, maps the influence of a region to nearby genes.

First, LISA constructs a null model of gene influence, which assumes each accessible region is occupied by its associated factors, and that all factor-bound regions exert influence on nearby genes. LISA then tests for the influence of a factor on a gene by calculating what proportion of that gene's influence could be attributed to that factor binding nearby regions. When you provide genes-of-interest, LISA finds factors that preferentially affects these genes over a sampling of background genes.

Refer to the User Guide to see it in action. Refer to the Data Analysis Guide to see the questions LISA can help you answer.

Requirements

Mac or Linux OS
Python 3.6+
15 GB of available storage space

Installation

LISA will install data into the virutal environment's "site_packages" directory, so ensure the env's location can store ~10GB.

PyPI

It is recommended to install lisa to a virtual environment:

$ python3 -m venv .venvs/lisa_env
$ source .venvs/lisa_env/bin/activate

Install LISA to this virtual env using this command:

(lisa_env) $ pip install lisa2

Conda

First, create a virtual environment:

(base) $ conda create --name lisa_env
(base) $ conda activate lisa_env

Then install from Conda:

(lisa_env) $ conda install -c liulab-dfci lisa2

Dataset Installation Issues

If you successfully install lisa but the program fails while downloading data, follow these manual dataset installation instructions.

Usage

Command Line Interface

LISA's cli offers convenient methods for the most common use cases. See the API, or try:

(lisa_env) $ lisa {command} --help

for parameter descriptions. See the User Guide for best practices.

Python Interface

The python module allows more control over the LISA test and more convenient data analysis. See the Python API and the User Guide.

Changelog

[2.2.4] - 2021-03-01

Added "lisa deseq" interface for parsing DESeq2 output files for fast LISA tests of DE genes

[2.2.0] - 2021-01-10

Added

Added "FromRegions" test, and moved all older functionalities to "FromGenes". New feature allows user to run LISA test with their own regions-of-interest
Added "query_reg_score" and "background_reg_score" matrices to output metadata of "FromRegions" test, which allows user to see which genes are likely regulated by each factor.
New backend interface for faster file transfers
Added ability to append more data to backend for future updates, including ATAC-seq epigenetic backgrounds
Added more documentation and user guide
Appended new ATAC data and reprocessed motifs using JASPAR database

Removed

Removed "cores" option from multi and oneshot tests, and removed mutliprocessing from package.
Removed "one-vs-rest" test because proved to provide unstable results

[2.1.0] - 2020-12-01

Bugfixes in output of "lisa multi" test
Refactored classes for future extension to user-supplied fragment files and peaks
Added integration testing
Added factor accessibility introspection to results printout
Made RP maps substitutable for future tests
Made assays modular so users can specify which statistical tests they are interested in

[2.0.6] - 2020-11-22

Support for Lisa version 1 API for integration with LISA website
Bugfixes in motif mode results
Slight speedups in parallelization of insilico-delition computing

Support

If you have questions, requests, or issues, please email alynch@ds.dfci.harvard.edu.

Wang-Cankun/lisa2