Metabolomics-centred graph-based multi-omics integration

mantra is a conceptual approach to compute estimates for the change of metabolic reaction activity between two groups of samples. It relies on linear relationships between substrate and product metabolites of a reaction and how the coefficients of these relationships change between conditions. In addition to analyzing metabolomics data, mantra also provides a correlation-based approach for multi-omics integration.

As an approach to provide smaller, mechanistically interpretable results based on both the reaction estimates (and multi-omics associations) and the metabolic network structure, network enrichment on the basis of a simulated-annealing assisted local search is used.

The pymantra package provides all functionalities for computing changes in reaction activity, multi-omics associations and performing the network enrichment as well as reporting and plotting their results. Additionally, it contains utilities to perform metabolite ID mapping.

```mermaid flowchart LR

subgraph in[Required Input] met(Metabolome Data) end

subgraph opt[Optional Multi-Omics Data] direction LR subgraph opti[Input] direction LR org(Organism/Microbiome Data) cat(Genome/Transcriptome Data) end subgraph optc[Associations] associ[computemultiomicsassociations] end end

subgraph db["Databases (online and/or offline)"] neo4j[(neo4j Graph)] sql[(SQL ID maps)] end

namemap[NameMapper
Metabolite ID Mapping] netgen[NetworkGenerator] metnet(((Metabolic Network)))

estimates[Reaction Estimates] lso[Local Search Enrichment] res[Enrichment Results] reac[ ]

sql -.-> namemap met --> namemap --> netgen neo4j -.-> netgen

opti --> netgen --> metnet

metnet & opti & estimates --> optc metnet & met --- reac ----> estimates

estimates & optc --> lso --> res

style reac width:0px,height:0px ```

Manuscript

If you would like to learn more about the details of the methodology and see some real-world results please check out our manuscript.

Getting Started

To get started we recommend going through some code examples provided in pymantra's documentation and the installation guide below.

Installation

Using pip

The recommended installation is via pip. It has pre-compiled wheels available for all major operating systems.

Generally using python version 3.10 or newer is recommended, but backwards compatability is ensured down to 3.8.

If you would like to run the experiments presented in the paper (code in the experiments folder), please use the second option to install the additional dependencies and mind the different version requirements (see README).

Additionally, dependencies required for package development and generation of sphinx documentation can be obtained.

```shell

install with package dependencies

pip install pymantra

for including the dependencies to run the paper experiments

pip install pymantra[experiments]

for including the development dependencies

pip install pymantra[dev]

for including the documentation dependencies

pip install pymantra[docs] ```

Installation from source

To install from source first clone the github repository including submodules shell git clone https://github.com/lipitum/pymantra.git --recursive

Additionally, you need to download required database files for name mapping from here

On macOS and Linux we recommend to use the following from within the pymantra directory

shell pip install gdown gdown --folder https://drive.google.com/drive/folders/1sWTHWRqzglWOjTMwNa_aqkKT__Y7Snmq?usp=sharing mv package/chebi.db pymantra/namemapping/ChEBI/ mv package/hmdb.db pymantra/namemapping/HMDB/ mv package/mantra.db pymantra/namemapping/mantra_db/ mv package/reactome.db pymantra/namemapping/Reactome/ rm -r package

On Windows the syntax would change to (untested)

pip install gdown gdown --folder https://drive.google.com/drive/folders/1sWTHWRqzglWOjTMwNa_aqkKT__Y7Snmq?usp=sharing move package\chebi.db pymantra\namemapping\ChEBI\ move package\hmdb.db pymantra\namemapping\HMDB\ move package\mantra.db pymantra\namemapping\mantra_db\ move package\reactome.db pymantra\namemapping\Reactome\ rmdir /s /q package

Make sure you have a C++ compiler installed (recommended are gcc for Linux and OS X and Visual C++ for Windows). Please make sure the compiler matches the one, with which your python distribution was installed.

The only c++ dependency for compilation is the boost library. In case it is already installed on your system (with version >= 1.77) you can also drop the --recursive flag in the git clone call. In addition to installing boost from source you can also use conda.

In addition to boost, we use OpenMP for parallel processing. It is usually installed with your compiler, although macOS users might have to run additional installation steps. Windows has currently no support for new OpenMP standards. If it is not possible to compile with OpenMP, parallelized options will be unavailable in the python package, but all other functionalities are not affected.

On macOS installation can be done on different ways, e.g. through MacPorts or homebrew (recommended - brew install libomp).

Once the c++ dependencies are installed, the package installation can be invoked by calling pip install shell pip install .

To verify installation run the unit tests shell python -m pytest

Similar to the regular pip installation you can also install the optional dependencies.

Database

Per default the setup script installs all code, but not the neo4j database. Instead, an online version of the database will be used. Since this is slower than having a local database, we provide two options to run the neo4j database locally.

Docker

We provide a docker script to run the neo4j database on your computer in a separate repository. Please follow the instructions there to run the container.

When calling the NetworkGenerator make sure your provide the correct URI and the user and password given in your .env file int pymantra-db-api folder. If you use the default setting this leads to python APINetworkGenerator("127.0.0.1:8084")

In this case only APINetworkGenerator.get_subgraph and APINetworkGenerator.as_networkx are available instead of the whole NewtorkGenerator function repertoire. Yet, for most use cases these functions are all you need.

Manual Installation

If you want to generate the networks outside docker, you will have to install neo4j on your systems (community version is sufficient). For instructions see the installation page.

Having installed neo4j, download the mantra database dump from here and move the

[comment]: <> (TODO: finish the instruction)

The NetworkGenerator can now be initialized with python NetworkGenerator("bolt://127.0.0.1:7687", ("<user>", "<password>"))

Manuscript Experiments

All code used to generate the results presented in our manuscript is located in the examples folder. Please mind the higher version requirements indicated in the README when reproducing the experiments.

Citation

If you use pymantra for your own work, please cite our manuscript

bibtex @article{koehler2023, title = {Identification and Integration of Key-Metabolic Reactions from Untargeted Metabolomics Data}, author = {K\"ohler, Nikolai and W\"urf, Vivian and Rose, Tim D and Pauling, Josch K}, journal = {bioRxiv}, year = {2023}, publisher = {Cold Spring Harbor Laboratory}, doi = {https://doi.org/10.1101/2023.05.15.540613} }