oxpig / ablang Goto Github PK

From the notebook, and running the model internally, it doesn't appear that the probability of an insertion is returned by the model.
Is that intentional?

For certain sequences, the most likely solution could be an insertion instead of an amino-acid.

How does one convert the residue likelihoods into probabilities? Is it by softmax operation?

See issue here: facebookresearch/fairseq#3535

Using Python 3.8.12 I was able to install ablang via pip, but Python 3.9 and 3.10 leads to the error described in the issue linked above.

Cool code trying it for a few things in the lab.

For the "Residue likelihood" I wanted to check. The first and last probabilitiy arrays are the SOS and EOS and do not correspond to a "real" amino acid residue?

Are the probabilities in alphabetical for the 20 essential amino acids?

Currently def seqcoding(self, seqs, align=False) only supports generation of res_codings using res_to_seq(mode="mean") in line:

Consider updating the the seqcoding method to also allow changing the aggregation mode of "sum", which is supported by res_to_seq.

Hi, The path to your https://opig.stats.ox.ac.uk/website/data/downloads/... gets 404'd, and as a result loading a pretrained model is impossible.

Hi there,

thanks for providing the code and model for use.

I'm having trouble replicating the example:

`import ablang

heavy_ablang = ablang.pretrained("heavy")
heavy_ablang.freeze()`

prints "Downloading model ..."

Which hangs there with no further logging for many hours.

I find I can navigate via web-browser to the url specified in the code:
"http://opig.stats.ox.ac.uk/website/data/downloads/" and download the tarball of the model there.

Best,
Michael

Hi, thanks for your work!

Can advise me the on data pre-processing pipeline that you used? Which OAS columns did you use to transcribe and translate DNAs to antibodies? How did you implement the transcription? In general the question is how did you go from raw OAS data to the data used in the paper. Thanks!

Would it be possible to release data-processing pipeline?