ECI-data

ECI stands for Enzyme-Chemical Interaction. The problem formulation is as below:

The high goal of ECI is to develop models which could use data of the activity of multiple compounds on related enzymes to enhance predictions of the activity of these compounds on the focused enzyme.

The problem formulation of ECI is similar to drug-target interaction(DTI), but there are some significant differences between the two: DTI is a matrix completion problem where the problem is information insufficiency. ECI is a matrix factorization problem where the problem is information overload.

The repository contain datasets suitable for ECI tasks come from the literature performing high-throughput experiments or large-scale characterization. It also contains the data curation codes and useful tools for biological and chemical data collation and text and image processing.

File Organization

Refer to the tree.txt for details.

directory	description
datasets	the pure datasets, with the four classical files listed in the Data Format part.
process	parse functions, the raw data files and the long datasets. In each datasets, the enzymes_long.csv and chemicals_long.csv are provided, if available. In the two files, the descriptors for enzymes and chemicals from the original paper are provided for reproducing the data analysis in the original paper.
util	utility functions for data processing.
media	Media files used in README.md.

Data Format

Four files are provided for each dataset.

file	description
enzymes.csv	Name, Sequence, pdb and other properties
chemicals.csv	Name, SMILES, sdf file path and other properties
activity.csv	raw values from literature
metadata.json	the original data collection strategy, modeling strategy, and enzyme catalysis issues to be explored

Metadata description

This repository not only focuses on the data itself, but also focuses on the scientific or engineering issues to be explored, the original data collection strategy and modeling strategy. The definition of each entry is listed in util\metadata.json . And the metadata.json in each dataset is edited manually after carefully reading the original paper.

The entries of meta data could be divided into five categories.

Type	Description
basic	Basic information of the literature, the enzymes family, the substrate library and the activity definition.
statistic	Statistics calculated from the data tables.
experiment	Experimental details of the enzymes' production and activity assay.
model	Model constructed in the original paper to rationalize the experimental results
process	data process implemented here

Datasets

Here are 7 datasets now. Run the code to make the statistical table below, as the statistic.csv file.

python main.py -s

Dataset	Description	Reaction	#E	#C	#A
esterase	A broad ester library to a broad esterase collections.		147	96	14112
hadsf	Broad phosphate to broad phosphatase		216	167	36072
olea	broad pNP substrate to broad OleA proteins		73	15	1095
nitrilase	12 bacterial nitrilases vs. 20 nitriles		12	20	240
bkace	DUF849 Pfam family beta-ketoacid lyase activity targeting beta-ketoacids		163	17	2771
fdh2	halogenase vs. mainly three types of compounds: indoles, anilines, phenols		38	62	2356
gt	glycosyltransferases from Arabidopsis thaliana vs. acceptors, α-Glucose-UDP		53	58	3074

Data Curation

We recommend that you install the python packages by conda.

conda create -n eci_data python=3.9
conda install --yes --file requirements

Active the conda environment

conda activate eci_data

Run the code to make a test.

python main.py -t [-v]

For a new dataset, run the code to generate the directory and parse_[dataset_name].py and metadata.json files.

python main.py -n [dataset_name] [-v]

After collecting the raw data files and editing the parse_[dataset_name].py file, run

python main.py -n [dataset_name][-v]

The main.py will call the parse() function in the parse_[dataset_name].py to curate the data. Feel free to change the code in parse_[dataset_name].py and output finally the desired files and use dev.ipynb file in the dir to test codes.

The operation of accessing the online server, such as submitting a task to Swiss-model, needs to wait for a period of time before accessing it to obtain content. Run the following command to update the table information and download files.

python main.py -n [dataset_name] -o [-v]

As a text file, one could edit metadata.json at any moment.

Related Sources

A very good beginning for ECI problem. A similar work based on different starting point:

[1] samgoldman97/enzyme-datasets: Enzyme datasets used to benchmark enzyme-substrate promiscuity models (github.com)

A very good introduction to DTI problem:

[2] alizat/Chemogenomic-DTI-Prediction-Methods: Algorithms for prediction of drug-target interactions via computational (chemogenomic) methods (github.com)