Hi!

I downloaded everything written in the README, than tried to run trDesign on a protein. However, I get the following

No such file or directory: '../../tools/trDesign/models/model_xaa.npy'

None of the downloaded files or the github repo contained a model_xaa.npy. How can I obtain it?

Thank you!

Hi, I tried to run by python design.py -p 1QYS.pdb -o tmp.txt but it appeared:
< ERROR: Output file not defined --out=None >

Is the license included in 01-hallucinate intended to apply to the whole of the trDesign repository, or is it intended for only 01-hallucinate?

I am trying to create a structure from the amino acid sequence using the tool.

Running this code gives me a somewhat linear feature space with little off diagonal elements. The predicted protein shape isn’t close to the one in the protein data bank.

seq = "MRYRLAWLLHSALPSTFRSVLGARLPPSERLCGFQKKTYSKMNNPAIKRIGHHIIKSHEDKREYRGLELANGIKVLLVSDPTTDKSSAALDVHIGSLSDPPNIAGLSHFCEHMLFLGTKKYPKENEYSQFLSEHAGSSNAFTSGEHTNYYFDVSHEHLEGALDRFAQFFLCPLFDESCKDREVNAVDSEHEKNVMNDAWRLFQLEKATGNPKHPFSKFGTGNKYTLETRPNQEGIDVRQELLKFHSTYYSSNLMAICVLGRESLDDLTNLVVKLFSEVENKNVPLPEFPEHPFQEEHLKQLYKIVPIKDIRNLYVTFPIPDLQKYYKSNPGHYLGHLIGHEGPGSLLSELKSKGWVNTLVGGQKEGARGFMFFIINVDLTEEGLLHVEDIILHMFQYIQKLRAEGPQEWVFQECKDLNAVAFRFKDKERPRGYTSKIAGILHYYPLEEVLTAEYLLEEFRPDLIEMVLDKLRPENVRVAIVSKSFEGKTDRTEEWYGTQYKQEAIPDEVIKKWQNADLNGKFKLPMKNEFIPTNFEILSLEKEATPYPSLIKDTAMSKLWFKQDDKFFLPKACLNFEFFSPFAYVDPLHCNMAYLYLELLKDSLNEYAYAAELAGLSYDLQNTIYGMYLSVKGYNDKQPILLKKIIEKMATFEIDEKRFEIIKEAYMRSLNNFRAEQPHQHAMYYLRLLMTEVAWTKDELKEALDDVTLPRLKAFIPQLLSRLHIEALLHGNITKQAALGIMQMVEDTLIEHAHTKPLLPSQLVRYREVQLPDRGWFVYQQRNEVHNNCGIEIYYQTDMQSTSENMFLELFCQIISEPCFNTLRTKEQLGYIVFSGPRRANGIQGLRFIIQSEKPPHYLESRVEAFLITMEKSIEDMTEEAFQKHIQALAIRRLDKPKKLSAECAKYWGEIISQQYNFDRDNIEVAYLKTLTKEDIIKFYKEMLAVDAPRRHKVSVHVLA" _seq = np.eye(20)[AA_to_N(seq)] preds = model.predict({"I":_seq[None,None]}) plot_feat(preds["feat"]) get_pdb(preds,"random.pdb")

Actual shape
https://www.uniprot.org/uniprotkb/Q24K02/feature-viewer

Hello,

I am running TrDesign_GD_demo.ipynb on Linux (not in Colab) and in Part 2 Step 2 I get many warnings:

WARNING, for demo we use n_models=1, in practice you should ALWAYS use n_models=5

model = mk_design_model(add_pdb=True, n_models=1)

mode: single sequence design
WARNING: Entity <bound method PSSM.call of <models.PSSM object at 0x2b5fa15e9b90>> could not be transformed and will be executed as-is. Please report this to the AutgoGraph team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output. Cause: converting <bound method PSSM.call of <models.PSSM object at 0x2b5fa15e9b90>>: AttributeError: module 'gast' has no attribute 'Num'
loading model: xaa
WARNING: Entity <bound method instance_norm.call of <models.instance_norm object at 0x2b5fa1648250>> could not be transformed and will be executed as-is. Please report this to the AutgoGraph team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output. Cause: converting <bound method instance_norm.call of <models.instance_norm object at 0x2b5fa1648250>>: AssertionError: Bad argument number for Name: 3, expecting 4
...

The warnings do not appear if I run the notebook in Colab. Also, after the warnings the does still run and would appear to generate correct results. Just thought you might want to know.

Thanks!

Scott

Hello,

I was wondering if it is possible to do a partial sequence design? What I'm thinking is that I would like to start with a known backbone (from a PDB file) and then mask out some of the residues and allow the model to predict only those.

I can see there is some functionality to 'spike' the seq_start but I think I'm thinking of something a bit different. Also, it would probably be necessary to backprop and update only the residues covered by the mask.

Like to hear your thoughts or ideas as I start thinking about this.

Thanks!

Scott

Does this code have a visual effect?

The process of reproducing the code has the following questions：

1. Is this code using supervised learning, semi-supervised learning or unsupervised learning?
2. Where is your training model and training data set?
   Looking forward to your recovery.