Tapioca, an integrative machine learning based computational pipeline for PPI networks in dynamics contexts. Tapioca integrates mass spectrometry (MS) curve data from thermal proximity coaggregation (TPCA), co-fractionation (CF-MS), or ion-based proteome-integrated solubility alteration (I-PISA), with protein physical properties, domains (PFAM, now InterPro), and tissue-specific functional networks (HumanBase). Tapioca itself consists of eight distinct logistic regression models that take unique combinations of these features as inputs. The predictions of these eight models are combined into a final interaction score for each pair of proteins in a given dataset.
The Tapioca website contains Tapioca scores for protein-protein interactions for thousands of proteins detected throughout cell culture infections with herpes simplex virus 1 (HSV-1; Justice et al. 2021), human cytomegalovirus (HCMV; Hashimoto et al. 2020) and Kaposi’s sarcoma associated herpesvirus (KSHV; unpublished). Individual or groups of proteins can be searched within these datasets on the Tapioca website, and the data can be downloaded.
To run Tapioca on your computer you will need both this repository and tissue specific functional networks from HumanBase (files are slightly too large to be uploaded to GitHub). The following subsections describe how to install and set up Tapioca locally, and use it to predict protein-protein interactions.
To install Tapioca locally you should first clone this repository as follows.
Before cloning the repository, make sure you are inside the directory in which you would like the repository to be placed.
git clone https://github.com/FunctionLab/tapioca.git
To navigate into the Tapioca folder use the command:
cd tapioca
We recommend managing Tapioca's dependencies with a conda environment as follows:
conda env create -f tapioca_env_mac.yml
If you are using a windows machine, use the "tapioca_env_windows.yml" file instead. If you wish to build the environment yourself, all you need is Python(Version >= 3.8) Numpy, Scipy, Pandas, Scikit-learn (Version 0.24.2), and Jupyter (if you wish to use jupyter notebook).
To activate the tapioca environment use the command:
conda activate Tapioca
Now you will need to download tissue specific functional networks tissue specific functional networks from HumanBase. On the linked HumanBase webpage you will see a list of networks
for different tissues. Simply download click the download button in the Top edges column for which ever network(s) you wish to use. If you are unsure which one to use, we recommend starting with the Global network. Be sure to place the downloaded files in the /data/tissue_functional_networks/ directory so that Tapioca can access them. Tapioca is capable of running without these networks. If you'd like to do so, then you can skip this text.
There are three main ways to run Tapioca. In either case, you will need to set a few variables such that Tapioca knows where your data is located, how to read it, and how you want Tapioca to run. These variables are:
This is the name of the csv file contain the data which Tapioca will run on. This cdv file should be located in the raw_input folder. To see how to format your input file, please see the example.csv file in the raw_input folder.
If you included a reference sample, and which this sample to be used in normalization, please set this variable to True. If you have no idea what a reference sample is, you should probably set this variable to False.
If you have already normalized your data, then this variable should be set to True, otherwise it should be False. If you are using co-fractionation data, we recommend normalizing your data first before inputing it into Tapioca. If this is set to True, be sure to place your data in the normalized_input folder, and see the example_normalized.csv file in the normalized_input folder for how to set up your file.
If you are using co_fractionation data, this variable should be set to True.
If you would like to run the full Tapioca model (all 8 sub-models and weighted integration) then set this variable to True. If you would like to run only the base model, which considers only the data in your input file and nothing else, set this variable to False.
Specific the file name of the tissue specific functional network you would like Tapioca to use. Remember you must download these networks from HumanBase yourself (see above). If you would like to not use any networks, set this variable to ''. Tapioca can only use one network at a time. If you have conditions that should be run in different tissue contexts then run Tapioca multiple times with different tissues, including only the conditions relevant to that tissue (in the conditions variable).
Set the name that you would like to have your output (Tapioca predictions) named as. The predictions will appear in the predictions folder.
These variables need to be set in 1 of 2 places, depending on how you would like to run Tapioca. If you would like to run Tapioca by running the main.py file, these variables should be set in the config.py file. If you would like to run Tapioca using jupyter notebook, these variables can be set within the Tapioca_Notebook.ipynb file. To open this file, first run jupyter notebooks by typing the following command in your conda terminal:
jupyter notebook
The jupyter notebook server should then open in your internet browser. Simply navigate to the tapioca folder and open the Tapioca_Notebook.ipynb file.
The third way to run Tapioca is via the command line. To accomplish this, first navigate to the 'tapioca' folder. Next type one of the following command:
run_tapioca.py -help
Or
python3 run_tapioca.py -help
Running this command will print out the symbols associated with each variable described above. Here, instead of setting relevant variable True and False, they are set 1 (True) and 0 (False). Here is of running the full model with a global tissue specific-functional network on the example file.
python3 run_tapioca.py -i example.csv -o cmd_savename -r 1 -f 1 -t global_top.gz
Two examples files are provided, both are located in the raw_input folder. One, called 'example.csv' is a very shortened version of the raw KSHV TPCA data produced in the Tapioca study. Running Tapioca on this file should only take a few minutes (per replicate, per condition) on most computers. The second example, 'full_example.csv', is the complete version of the raw KSHV TPCA data. Running Tapioca on this file should take around 8 hours (per replicate, per condition).
Tapioca has recently been submitted for publication.