dbase_prep

Prepare a compound database for substructure and similarity searches.

The workflow described in this repository starts with a compound database in SMILES format. Using RDkit, fingerprints for substructure as well as similarity searches will be calculated.

Requirements

• Compound database in SMILES format
• RDkit
• Numpy
• Pandas
• Scipy
• Glob (for chunked workflow)

Usage

For larger compound sets (upwards of 6 M compounds), this workflow might fail due to insufficient memory, and the use of a proper database, such as the RDkit database cartridge, is recommended. I haven't had the opportunity to play with the cartridge yet, so as a workaround, calculation of Morgan Fingerprints can be performed in chunks.

An example file with 58 compounds is provided (58_smiles.smi).

Fingerprinting In One File

1. Convert SMILES into RDkit mol object: python smiles2molobj.py

   Define smiles_name, pickle_name, failed_name in script

2. Calculate Morgan Fingerprints as Bit Vectors: python molobj2morganbv.py

   Define pickle_in, bv_pickle_out, radius in script

3. Convert Morgan FP Bit Vectors into Scipy Sparse Matrices: bv2csr.py

   Define file in script

Fingerprinting In Chunks

1. Convert SMILES into RDkit mol object: python smiles2molobj.py

   Define smiles_name, pickle_name, failed_name in script

2. Calculate Morgan Fingerprints as Bit Vectors: python molobj2morganbv.py

   Define pickle_in, bv_pickle_out, radius in script

3. Split the Morgan bv pickle into chunks: python splitmorganbv.py

   Define bv_pickle_in, n_chunks in script

4. Convert Morgan FP Bit Vectors into Scipy Sparse Matrices: python bv2csr_chunks.py

   Define path, files in script

5. Concatenate the chunks: python concatchunks.py

   Define path, file, giant_pickle in script

To-Do

• Improve UI: move arguments from inside scripts to proper command line arguments