sigRecon: Signature Recontextualization

The signature recontextualization problem describes a simple goal in computational biology: Given a gene signature X of a genetic or chemical perturbation in model organism Y, what is the corresponding gene signature of the same perturbation in model organism Z?

Any method that is able to input a ranked list of genes from one biological context and output another ranked list of genes for another context performs this task of 'signature recontextualization'.

This repository contains benchmarking tasks and data for evaluating signature recontextualization (as reported in sigrecon_benchmarking), plus implementations of two of our own methods: projection-based scoring (projectCor()) and network propagation (netProp()).

Installation

Note: BiocManager installer is used to handle dependences. This package is currently not hosted on Bioconductor.

BiocManager::install("montilab/sigrecon", dependencies = TRUE)

Quick Start

The example below runs entirely on a small, bundled real-data example: demo_sciplex_sigs (source-context signatures), demo_sciplex_se (target-context expression), and demo_sciplex_true_sigs (target-context ground truth). Demo data is bundled for each dataset in the benchmarking study (Perturb-seq, SciPlex, DrugMatrix, Tahoe); the full pseudobulk expression and perturbational signatures for each are archived on Zenodo (see Data below).

library(sigrecon)

# Recontextualize with projectCor
recon_projectcor <- projectCor(demo_sciplex_se, demo_sciplex_sigs, score = "gsva")

# Or with network propagation
recon_netprop <- netProp(demo_sciplex_se, seeds = demo_sciplex_sigs, sig = "rwr")

# Benchmark against the true target-context signature
eval_df <- sig_eval_table(
  source_sigs = demo_sciplex_sigs,
  pred_sigs   = recon_projectcor,
  true_sigs   = demo_sciplex_true_sigs,
  source      = "sciplex_k562",
  target      = "sciplex_a549"
)
head(eval_df[, c("gene", "jacc", "NES", "padj")])

Data

Full-size pseudobulk expression and perturbational signatures for each dataset in the benchmarking study are archived on Zenodo:

Dataset	Perturbational signatures	Pseudobulk expression
DrugMatrix	10.5281/zenodo.21432933	10.5281/zenodo.21433031
SciPlex	10.5281/zenodo.21432935	10.5281/zenodo.21433011
Perturb-seq	10.5281/zenodo.21432937	10.5281/zenodo.21433138
Tahoe	10.5281/zenodo.21433000	10.5281/zenodo.21433050

The small, bundled demo_* datasets used above are subsets of these; use get_dataset()/list_datasets() for the full processed signature sets, or the Zenodo records above for the underlying pseudobulk expression.