BiGG Integration Tool Analysis

Multivariate analysis of Genome-Scale Metabolic Models (GSMMs) derived from BiGG Models database.

Workflow

1. merlin 's BiGG Integration Tool was used to derive GSMMs from BiGG Models database. BiGG Integration Tool allows one to reconstruct draft GSMMs from BiGG models using BLAST similarity searches. A draft reconstruction can be performed using as many BiGG template-models as you would like. In this work, we have performed several draft reconstructions for three organisms, namely Streptococcus thermophilus, Xylella fastidiosa, and Mycobacterium tuberculosis. The draft GSMMs have been reconstructed using the following templates:

   • whole BiGG database;
   • a set of functionally close models. Close models have been inferred from the COG metabolic functional annotation (see bellow);
   • a set of randomly selected models (this procedure was repeated 5 times).

2. In addition, carveme was also used to generate a GSMM for each organism using the BiGG Models database.

3. The multivariate analysis performed for the resulting models can be consulted in this repository.

Analysis

Comparative Genome Functional Analysis

1. NCBI's RefSeq database was used to retrieve the proteome sequence of each organism represented in the BiGG Models database.

2. A COG functional annotation was performed for each organism using reCOGnizer. Then, COG metabolic functional annotation was used for further analysis.

3. Metabolic COG identifiers were used to perform a Principal Component Analysis (PCA). For that, a binary matrix was constructed in which each row represented an organism of BiGG database and each column represented a COG identifier. All matrix cells were filled with presence (1) or absence (0) of the respective COG identifier. Variables having low variance were filtered out and standardization was applied, using VarianceThreshold and StandardScaler methods from sckit-learn, respectively. A PCA analysis was performed in this matrix using also sckit-learn.

4. The analysis of variance for the reCOGnizer metabolic functional annotation was then compared with each organism domain and phylum.

Model Analysis

1. Relations among models have been investigated using Venn diagrams and PCA analysis.

2. COBRApy package has been used to read the draft models. Reactions, metabolites, and genes shared across the draft models have been inferred. Furthermore, COG functional annotation of all models' genes have also been used to assess the draft models' similarity.

3. matplotlib_venn package has been used to visualize shared reactions, metabolites, genes and COGs.

4. sckit-learn and matplotlib packages have been used to run and plot multivariate PCA analysis over reactions, metabolites and metabolic COG identifiers associated to the models' genes.

Repository

• genomes_analysis: results of the functional analysis performed to the organisms available in BiGG Models.
• model_analysis: results of the multivariate analysis performed to the draft models.
• models: draft models and some statistics
• resources: resources for the functional analysis performed to the organisms available in BiGG Models.
• cog_analysis: module with functions to run the COG functional annotation using reCOGnizer
• genome_resources: module with several tools to obtain RefSeq proteome sequences for all organisms represented in BiGG Models database.
• pca_analysis: module with functions to run the PCA analysis over the models' reactions, metabolites and COG annotation
• random_models: module to select random BiGG models as template for the three organisms.
• utils: utilities
• venn_diagram: module with functions to run Venn's diagrams over the models' reactions, metabolites and COG annotation

Requirements

• merlin and BiGG Integration Tool - draft models
• carveme - draft models
• reCOGnizer - COG functional annotation
• Biopython - proteome sequences retrieval
• COBRApy - read models
• pandas - data manipulation
• sckit-learn - PCA analysis
• matplotlib - plot
• matplotlib_venn - plot

Installation

Clone this repository: git clone https://github.com/BioSystemsUM/bit-analysis.git Consult the previous requirements list to set up all mandatory packages and tools.

Utilization

BIT can be used directly on merlin (https://merlin-sysbio.org/). Alternatively, it can be used programatically using the python script available in the "BIT tool files" folder - modelGenerator.py or with the command:

java -jar -Xmx5G -Xss512M -XX:+HeapDumpOnOutOfMemoryError -Djavax.xml.accessExternalDTD=all -Dworkdir=workdir/ Blast.jar " + params)

Params:

option

, where option can have the following values: 1 - Use the complete BiGG database 2 - Select specific models as template 3 - Use three random models as template

template_models

the name of the models to use as template

eValue, bitScore, queryCoverage

blast parameters for e-value, bit-score, and query coverage

includeReactionsWithoutGPRBigg, includeReactionsWithoutGPR

Boolean parameters to allow including reactions with incomplete or without GPR in the BiGG database, or not.