MELLODDY - Public Data Preparation

Work files for the preparation of public data used to test the federative learning pipelines.

1. Extraction

The starter CSV is extracted from the ChEMBL25 database by running a SQL query. The relevant files can be found in the extraction directory.

2. Processing and Formatting

The notebooks and helper python modules in the processing directory perform:

• assay filtering to satisfy the selection criteria,
• assay fusion (statistical for binding and functional assays, type-based for physicochemical assays),
• binning of continues values into categories,
• computation of descriptive statistics for the resulting dataset,
• formatting of the final dataset into T2, T3, and T4 files,
• building a reference for assay IDs in case the original (pre-fusion) IDs are needed.

3. Data Files

Raw data extracts and processed files can be found here: https://zenodo.org/record/5045055

The archive includes:

• the T2, T3, and T4 files required for the federative learning pipeline,
• the initial dataset chembl_extract.csv extracted from the ChEMBL25 database,
• the assay ID reference assay_reference.csv,
• summary statistics on the final dataset in the README file.

4. Related packages

The extracted an processed output files can be used to to prepare files with MELLODDY-TUNER (https://github.com/melloddy/MELLODDY-TUNER) as inputs for SparseChem (https://github.com/melloddy/SparseChem).