smizip 2.0.0

An implementation of Roger Sayle's SmiZip algorithm for compressing short strings

SmiZip is a compression method for short strings. It was developed in 1998 by Roger Sayle (while at Metaphorics LLC) to compress SMILES strings, and fully described at a Daylight Mug01 presentation in 2001: https://www.daylight.com/meetings/mug01/Sayle/SmiZip/index.htm

This repo is an implementation Noel O’Boyle of the SmiZip algorithm in Python. This work was presented at the 12th RDKit UGM in Mainz in Sep 2023: https://github.com/SoseiHeptares/presentations/blob/main/2023/2023-09-12thRDKitUGM_NoelOBoyle_SmiZip.pdf

Note that the more recent ‘smaz’ (https://github.com/antirez/smaz) short string compression algorithm (2009) is equivalent in concept, but favours a greedy approach over an optimal encoding. This was tweaked for SMILES by Andrew Dalke as ‘smilez’; a Python 3 port by David Lorenzana can be found at https://github.com/davidlorenzana/smilez. Another more recent paper describes ZSMILES (2024, https://arxiv.org/abs/2404.19391), which seems very similar.

Quick start

Install as follows:

pip install smizip

First, let’s download a set of n-grams trained on RDKit canonical SMILES from ChEMBL:

curl https://raw.githubusercontent.com/SoseiHeptares/smizip/main/example-ngrams/rdkit.slow.json -o rdkit.slow.json

Now let’s use this to compress and decompress a .smi file that contains canonical SMILES from RDKit:

smizip    -i test.smi  -o test.smiz  -n rdkit.slow.json
smizip -d -i test.smiz -o test.2.smi -n rdkit.slow.json

Note that the expected format of a .smi file is [SMILES][tab][TITLE]\n.

Other example sets of n-grams are available from the GitHub site (https://github.com/SoseiHeptares/smizip/tree/main/example-ngrams). To create your own JSON file of n-grams, you can train on a dataset (find_best_ngrams), or modify an existing JSON (add_char_to_json).

To use from Python:

import json
from smizip import SmiZip

json_file = "rdkit.slow.json"
with open(json_file) as inp:
   ngrams = json.load(inp)['ngrams']

zipper = SmiZip(ngrams)
zipped = zipper.zip("c1ccccc1C(=O)Cl") # gives bytes
unzipped = zipper.unzip(zipped)

Note

You should include \n (carraige-return) as a single-character n-gram if you intend to store the zipped representation in a file with lines terminated by \n. Otherwise, the byte value of \n will be assigned to a multi-gram, and zipped SMILES will be generated containing \n.

A similar warning goes for any SMILES termination character in a file. If you expect to store zipped SMILES that terminate in a TAB or SPACE character, you should add these characters as single-character n-grams. Otherwise the zipped representation may contain these and you won’t know which TABs are terminations and which are part of the representation.

Changes

v2.0 (2025-01) : Python API: unchanged. find_best_ngrams.py: new option --non-printable to facilitate encoding into printable ASCII charactersr; --chars is now required (help text provides a reasonable starting point) to force the user to consider the list; if the end of the training .SMI file is reached, the script wraps around to the start; --cr corrected to --lf. compress.py: A better error message is generated if an attempt is made to encode a character not present in the JSON file; support added for .SMI files without titles. Thanks to Adriano Rutz (@adafede) and Charles Tapley Hoyt (@cthoyt) for feedback.