Preserving the dot symbol (".") in the alphabet

[vandrw]

Hi! The current version of the library provides the get_alphabet_from_selfies function that generates a list of symbols from an iterable. However, if one uses the returned alphabet to generate an encoding (as seen below), the script would raise an error that there is no symbol "." in the alphabet.

sf.selfies_to_encoding(
        selfies_string,
        vocab_stoi=symbol_to_index_dict,
        pad_to_len=16,
        enc_type="label",
)

After having a closer look at the functions involved, there are two possible solutions:

1. Preserve the dot symbol in the alphabet by removing the following line:
   

   selfies/selfies/utils/selfies_utils.py

   Line 71 in 120b776

   alphabet.discard(".")

2. Clean up the dot symbol from the char_list created here (i.e., char_list.remove(".")):
   

   selfies/selfies/utils/encoding_utils.py

   Line 50 in 120b776

   char_list = split_selfies(selfies)

In the context of training a machine learning model, would this symbol provide vital information? If so, wouldn't it be better to preserve it in the alphabet?

[MarioKrenn6240]

Hi @vandrw -- Thanks for your interest. The dot is used to represent two unconnected molecules, or unphysical bonds that cannot be represented in a native way within SMILES (Ferrocenes). For generative models, we didnt encounter any usecase of it yet. We only introduced it to be able to read even some il-formed SMILES. Hope this helps!