---
tags:
- chemistry
size_categories:
- 1B<n<10B
configs:
- config_name: default
  data_files: "zinc-druglike-cano.tar.xz"
---

Downloaded Nov 25th 2023 from ZINC-20 (https://zinc20.docking.org/).

Contains the unique canonicalized SMILES molecules, done with RDKit.

Used this filtering criteria in the ZINC-20 tranches:
- 2D representation
- Standard reaction
- Annotated purchasability
- Drug-like subset

Distribution of tokenized SMILES sequence lengths below, along with the top 15 SMILES tokens (with their frequencies). The following regex string was used 
to split the SMILES molecule into tokens: (\[[^\]]+]|Br?|Cl?|N|O|S|P|F|I|b|c|n|o|s|p|\(|\)|\.|=|#|-|\+|\\|\/|:|~|@|\?|>>?|\*|\$|\%[0-9]{2}|[0-9])

![Distribution of Sequence Lengths](sequence_lengths_distribution.png)
![Top 15 SMILES Tokens and Their Frequencies](top_smiles_tokens.png)