Question regarding the output of selfies

Hello,

I have a question regarding the output selfies. I trained an simple teacher-forced RNN based on selfies of molecules. My trained model generates selfies like the example below:

 [C][=S@@expl][=SHexpl][=P][/S+expl][/C@@Hexpl][/P@@expl][\Siexpl][C][C][C][/I][IH2expl]<sos>[C][SnH2expl][Sn+expl][=S@@expl][\S][#N+expl][=I][\I][C][C@Hexpl][Expl/Ring2][P@expl][C][\CH-expl][Branch1_3][Expl=Ring2][C][/B][P][\S][\C@Hexpl][=P@expl][Br+expl][=P][\S+expl][/O][/P][=S@@expl][N+expl][=P][/N][B-expl][=S][N@@+expl][Cexpl][\C@@Hexpl][/N][=N][Oexpl][\O][/C][/NHexpl][/S@expl][B-expl][S@@Hexpl][\S][C@@Hexpl][Sn+expl][SiH3expl][C][\B][S@@expl][Ring2][/C][P+expl][N][C][N][C][C][C][Branch2_3][=S+expl][BH2-expl][=Nexpl][#N][/C@@expl][Nexpl][CH-expl][/B][C][C][\N][S@@+expl][/P@expl][SnH4+2expl][Branch2_2][=P][/C@expl][/S+expl][S@Hexpl][\B][/C][N][Snexpl][BH3-expl][B-expl][\S][=S][N+expl][Sn+3expl][/I][\P][\S+expl][F][Cexpl][/S@expl][Siexpl][B@@-expl][/C@Hexpl][\Cl][/P@expl][O][O-expl][\C@@expl][C-expl][/O+expl][\O-expl][Branch1_3][N][/Cexpl][Expl\Ring1][B@-expl][O][N][SiH3expl][C][SnH6+3expl][=S@@expl][\S@expl][CH-expl][#N+expl][\S@@expl][/Cexpl][SnH4+2expl][=O][S@expl][O][\P][=Snexpl][Cexpl][C][/N+expl][P@@expl][/CH-expl][=IH2expl][N][=P@expl][#C][P@@Hexpl][P+expl][Br][=S@@expl][C-expl][/Br][=N+expl][P][=Siexpl][P][NHexpl][/Br][N+expl][\N+expl][N+expl][O][C][/C][\O-expl][=S@@expl][\S][BH-expl][Branch1_2][SnH2expl][Ring2][/CH-expl][=17Oexpl][\C@expl][=P@Hexpl][=P+expl][O][C][C][NHexpl][C][C][C][C][/P@@expl][N][B@@-expl][C+expl][=S][=P+expl][C][B][\I][=C][C][C][C][O][N][O][Branch2_1][SiH3expl][CHexpl][\S@@expl][/B][=P@Hexpl][Branch2_3][C@@Hexpl][/F][S@+expl][\N][/C][/B][Oexpl][P@@Hexpl][/NHexpl][/O][O][=SHexpl][\Oexpl][=Siexpl][P@+expl][SiH3expl][SnH6+3expl][O-expl][Branch1_2][CH2expl][\S@expl][=SHexpl][PHexpl][\F][\C@@Hexpl][O][C][O][Nexpl][O][C][N][Cl][N@@+expl][P@@Hexpl][O][C][/Cexpl][S@@expl][C][\O][\Br][/O][\N+expl][SHexpl][\Siexpl][=SHexpl][B@-expl][S][Br][\F][\S+expl][C][C][C][=N-expl][BH3-expl][#C-expl][/Cl][\S+expl][Expl=Ring2][CHexpl][C][Expl/Ring2][P@+expl][N-expl][#N][Br][\Snexpl][BH-expl][#C-expl][CH2expl][=Snexpl][/Cl][/P@expl][C][\Oexpl][Expl=Ring1]

However, after passing this SELFIES to sf.decoder() function, the output SMILES looks significantly shorter:

 C=[S@@]=[SH]=P/[S+]/[C@@H]/[P@@]\[Si]CCC/I

When I try to convert the SMILES back to SELFIES back again, I got different SELFIES:

[C][=S@@expl][=SHexpl][=P][/S+expl][/C@@Hexpl][/P@@expl][\Siexpl][C][C][C][/I]

Can I get some explanation why this happens? Thank you so much!

[shiwentao00]

Thank you so much for the detailed explanation. Final question: is github the right place to raise questions about using SELFIES? I may come back in the future with other questions.

[MarioKrenn6240]

Yes, github is perfect for such questions, it might help others with similar questions. Please keep them coming.