ERROR: failed to load model

[zbekkhozhin]

In the README, it is written to use:

cd network
wget https://files.ipd.uw.edu/dimaio/RF2_jan24.tgz
tar xvfz RF2_jan24.tgz
cd ..

But in the run_RF2.sh file, there is:

python $PIPEDIR/network/predict.py \
    -inputs $argstring \
    -prefix $WDIR/models/model \
    -model $PIPEDIR/network/weights/RF2_apr23.pt \
    -db $HHDB \
    -symm $symm #2> $WDIR/log/network.stderr 1> $WDIR/

You can see that the namings do not match, the weights are inside a subdirectory and named differently and are actually themselves different. So use

cd network
wget https://files.ipd.uw.edu/dimaio/RF2_apr23.tgz
tar xvfz RF2_apr23.tgz
cd ..

instead. This will place the model weights into a correct folder from which RoseTTAFold2 can read in the model weights

There are some angry comments about the code being a train wreck, and we all understand that developers have to spend an immense amount of effort to maintain their code with all things getting updated and those dependencies ruining everyone's lifes, so these tiny mistakes in the wget might get past your eyes.

And the model inference on a gpu is super fast, cheers! Although the backbone in the unstructured regions has very strange geometry, the peptide bonds appear too long sometimes