braid

Find protein complexes from high-throughput AP-MS data

Python implementation for Markov Random Field model of AP-MS experiments.

Rungsarityotin, W., Krause, R., Schoedl, A., Schliep, A. (2007) BMC Bioinformatics 8:482.

Requirement

Require a Python >= 3.5 installation. Since it depends on some packages that can be tricky to install using pip (numba, numpy, ...), we recommend using the Anaconda Python distribution. In case you are creating a new conda environment or using miniconda, please make sure to run conda install anaconda before running pip, or otherwise the required packages will not be present.

In addition if you obtain the code from the development branch, we now use Tensorflow when GPU is available. The develop branch contains an implementation for large dataset like BioPlex 2.0 or BioPlex 3.0. We now have good results runnning on datasets with fewer than 4,000 proteins as well as on large human datasets with more than 10,000 proteins.

Input file

(-f) a CSV file containing a list of bait-preys experiments. The first protein at the beginning of every line is a bait protein. The option (-bp) must be given if your input file is Bioplex 2.0 or Bioplex 3.0.

Model parameters

The log-ratio of false negative rate and false positive rate (-psi)
Maximum possible number of clusters (-k), e.g. 600 for gavin2002 dataset and 800 for gavin2006

Example

py main.py -f gavin2002.csv -psi 3.4 -k 500 or py main.py -bp [Bioplex file] -psi 3.4 -k 500

Output

out.tab: a tab separating file containing a cluster annotation for each protein, one per line.

out.sif: output clustering result in the Simple Interaction File format which you can import into Cytoscape.

Comparison with ClusterONE

Benchmark: S.cerevisiae (MIPS), E.coli (EcoCyC)

Ref 1 (proteins from Gavin2006 PPI network):

MIPS	MRF	ClusterONE	Mix.Bernou.
Accuracy	0.504	0.498	0.450
Fraction	0.643	0.713	0.713

Note: Best MRF runs with parameters $\psi = 3.4$, $k \in [400, 500]$.

Ref 2 (proteins from Gavin2006 PPI network):

MIPS	MRF (w prior)	ClusterONE	Mix.Bernou.
Accuracy	0.495 (0.499)	0.498	0.450
Fraction	0.730 (0.713)	0.713	0.713

Ref 3 (MIPS proteins):

MIPS	MRF	ClusterONE	Mix.Bernou.
Accuracy	0.396 (0.384)	0.378	0.349
Fraction	0.492 (0.470)	0.428	0.433

Note: Best MRF runs with parameters $\psi = 2.4, 3.4$, $k \in [700,800,900,1000]$. ClusterONE: default parameters. Mixture of Bernoulli: $k=800$.

Ref 4 (proteins from Babu2018 PPI network)

EcoCyc	MRF	ClusterONE	Mix.Bernou.
Accuracy	0.457 (0.465)	0.558	0.454
Fraction	0.386 (0.397)	0.170	0.159

Ref 5 (EcoCyC proteins)

EcoCyC	MRF	ClusterONE	Mix.Bernou.
Accuracy	0.411 (0.372)	0.318	0.467
Fraction	0.157 (0.121)	0.032	0.186

Note: parameters of MRF $\psi = 0.1$, $k = 800$.