A tutorial to help out first-time users

[alptaciroglu]

Dear developers,

Thank you very much for developing Tranception. I am quite eager to try it out but as a first-time user I struggled running it. I think a tutorial with several example runs would help out users very much! (Apologies if there is one and I didn't see it.)

[pascalnotin]

Hi @alptaciroglu -- thanks for your interest in Tranception! We had put together this brief tutorial for TranceptEVE in the ProteinGym repo, which you may find helpful for Tranception as well. Let us know your thoughts / what would be helpful to include in a similar notebook for Tranception.

[nathan2v]

Hello, I am trying to figure out how to use tranception for single input analysis. I have managed to install it witout much problem but the examples show there needs to be a msa of a2m. I currently have a fasta that I want to run though, is there a special variable for single input sequences? I can't find it in the --help menu for score_tranception_proteingym.py.

[pascalnotin]

Hi @nathan2v -- by default Tranception will run on single sequence input only (no MSA retrieval). Example scripts from the ProteinGym codebase:

• single sequence input / no MSA retrieval: https://github.com/OATML-Markslab/ProteinGym/blob/main/scripts/scoring_DMS_zero_shot/scoring_Tranception_substitutions_no_retrieval.sh
• MSA retrieval (requires the --inference_time_retrieval flag): https://github.com/OATML-Markslab/ProteinGym/blob/main/scripts/scoring_DMS_zero_shot/scoring_Tranception_substitutions.sh

[nathan2v]

@pascalnotin Thankyou for your help. Is there a way to run without a DMS? My input is a fasta and a list of point mutations (P10L, A25T,Y50F, ect..). Or is there a way to generate or lookup an appropriate DMS file for the program to run?

My current goal is to run and compare against SIFT to benchmark traditional and AI versions and I was recommended to use this software as one of the programs to produce results with.

Currently been running something like this with a generated file I made using python but the scores are bad and I don't think it's correct:

python3 score_tranception_proteingym.py \
--checkpoint ./Tranception_Large \
--target_seq ./msafolder/ecoli_gyrA.faa \
--DMS_data_folder ./ \
 --DMS_file_name ../../Fasta_files/fasta_input/subst/gyrA.csv \
--output_scores_folder ./output_scores 

My DMS file is just two columns one with the mutation I want to test and the second with the full mutated sequence. The scoring however is really bad and I think I am missing something. SIFT had a .subst file that it used as input so I initially thought using the cvs as something similar might work.

I am currently unfamiliar with DMS files or if It is possible to get a DMS from a single FASTA and a sequence of Substitutions. From what I can tell most tools seem to require raw fragmented sequences that are then scored. However, I do not have raw data to use. And it seems, when reading on DMS, that the standards can change from software so I am unsure what Tranception actually needs to run properly in the DMS .cvs file.

Any pointers would be appreciated.

gyrA.csv