Gene GEM Correlation Analysis (GGCA)

Computes efficiently the correlation (Pearson, Spearman or Kendall) and the p-value (two-sided) between all the pairs from two datasets. It also supports CpG Site IDs.

IMPORTANT: GGCA is the heart of a platform called Multiomix. On the official website you will be able to use this library in a fast and agile way through a friendly graphical interface (along with many extra features!). Go to https://multiomix.org/ to get started now!

Python PyPi | Rust Crate

Index

• Requirements
• Usage
  • Python
  • Rust
• Contributing
• Considerations

Usage

There are a few examples in examples folder for both languages.

Python

1. Install: pip install ggca
2. Configure and call the correlate method:

import ggca


mrna_file_path = "mrna.csv"
gem_file_path = "mirna.csv"

try:
	(result_combinations, evaluated_combinations) = ggca.correlate(
		mrna_file_path,
		gem_file_path,
		correlation_method=ggca.CorrelationMethod.Pearson,
		correlation_threshold=0.5,
		sort_buf_size=2_000_000,
		adjustment_method=ggca.AdjustmentMethod.BenjaminiHochberg,
		all_vs_all=True,
		gem_contains_cpg=False,
		collect_gem_dataset=None,
		keep_top_n=2  # Keeps only top 2 elements
	)

	print(f'Number of resulting combinations: {len(result_combinations)} of {evaluated_combinations} evaluated combinations')
	for combination in result_combinations:
		print(
			combination.gene,
			combination.gem,
			combination.correlation,
			combination.p_value,
			combination.adjusted_p_value
		)
except ggca.GGCADiffSamplesLength as ex:
	print('Raised GGCADiffSamplesLength:', ex)
except ggca.GGCADiffSamples as ex:
	print('Raised GGCADiffSamples:', ex)
except ggca.InvalidCorrelationMethod as ex:
	print('Raised InvalidCorrelationMethod:', ex)
except ggca.InvalidAdjustmentMethod as ex:
	print('Raised InvalidAdjustmentMethod:', ex)
except ggca.GGCAError as ex:
	print('Raised GGCAError:', ex)

Rust

1. Add crate to Cargo.toml: ggca = { version = "1.0.0", default-features = false }
2. Create an analysis and run it:

use ggca::adjustment::AdjustmentMethod;
use ggca::analysis::Analysis;
use ggca::correlation::CorrelationMethod;

// File's paths
let df1_path = "mrna.csv";
let df2_path = "mirna.csv";

// Some parameters
let gem_contains_cpg = false;
let is_all_vs_all = true;
let keep_top_n = Some(10); // Keeps the top 10 of correlation (sorting by abs values)
let collect_gem_dataset = None; // Better performance. Keep small GEM files in memory

let analysis = Analysis::new_from_files(df1_path.to_string(), df2_path.to_string(), false);
let (result, number_of_elements_evaluated) = analysis.compute(
	CorrelationMethod::Pearson,
	0.7,
	2_000_000,
	AdjustmentMethod::BenjaminiHochberg,
	is_all_vs_all,
	collect_gem_dataset,
	keep_top_n,
)?;

println!("Number of elements -> {} of {} combinations evaluated", result.len(), number_of_elements_evaluated);

for cor_p_value in result.iter() {
	println!("{}", cor_p_value);
}

Note that env_logger crate is used to provide some warning in case some mRNA/GEM combinations produce NaN values (for instance, because the input array has 0 std). In that case, you can add RUST_LOG=warn to your command to produce warnings in the stderr. E.g:

RUST_LOG=warn cargo test --tests

or

RUST_LOG=warn cargo run --example basic

Development and contributions

All kind of help is welcome! Feel free o submit an issue or a PR.

• Build for rust: cargo build [--release] or run an example in the examples folder with cargo run --example [name of the example]
• Build and run in Python: run cargo build [--release] and follow the official instructions to import the compiled library in your Python script.
• Build for Python (uses Maturin) and it's generated by CI maturin-actions

Tests

All the correlation, p-values and adjusted p-values were taken from cor.test and p.adjust functions from the R programming language and statsmodels package for Python language.

Data in small_files folder was retrieved with random sampling from the Colorectal Adenocarcinoma (TCGA, Nature 2012) dataset. This dataset can be downloaded from cBioPortal datasets page or this direct link.

All the correlations results were compared directly with R-Multiomics output (old version of multiomix.org only available for R lang).

Performance

We use criterion.rs to perform benchmarks. In case you have made a contribution you can check that no regression was added to the project. Just run cargo bench before and after your changes to perform a statistical analysis of performance.

Considerations

If you use any part of our code, or the tool itself is useful for your research, please consider citing:

@article{camele2022multiomix,
  title={Multiomix: a cloud-based platform to infer cancer genomic and epigenomic events associated with gene expression modulation},
  author={Camele, Genaro and Menazzi, Sebastian and Chanfreau, Hern{\'a}n and Marraco, Agustin and Hasperu{\'e}, Waldo and Butti, Matias D and Abba, Martin C},
  journal={Bioinformatics},
  volume={38},
  number={3},
  pages={866--868},
  year={2022},
  publisher={Oxford University Press}
}