MPNN output explanation

[dhiraj82]

Hi
I am trying to design sequence using solubleMPNN and getting two type of output, one .fa file and another .cif file. fa file looks quite different to the one designed by old MPNN (as shown in its example) or dl_binder_design output. only number I am getting in the output header from foundry mpnn is sequence recovery. Also, it seems like the foundry output .fa file is reporting the sequence of binders as well as target combined while the one from dl_binder_design is reporting the sequence for binders only.
The cif file output from foundry mpnn cannot be visualized using pymol.
Is this normal? How can I proceed for the next step of prediction of complex with this output?
my script for running mpnn is written below. I tried to add fixed_chain option as in the original mpnn, but I am getting the error that designed chains and fixed chains option cannot be used together.

# Set the directory path to where your CIFs are:
PDB_DIR_PATH="./renamed/"

cif_files=($PDB_DIR_PATH/*.cif)

# Iterate and execute command on each file
for file in "${cif_files[@]}"; do
    mpnn --model_type "protein_mpnn" \
    --structure_path "$file" \
    --out_directory MPNN \
    --batch_size 2 \
    --number_of_batches 1 \
    --omit '["CYS", "UNK"]' \
    --designed_chains "A"  \
    --is_legacy_weights "True" \
    --checkpoint_path "/opt/checkpoints/solublempnn_v_48_020.pt"
done

[AndrewKubaney]

Hi @dhiraj82,

We are working on fixing the fasta output to have better support for multiple sequences. To match the original LigandMPNN repository, we plan on outputting both designed and fixed (target) residues. We are planning to include an entry for the input sequence (pre-MPNN) and headers with more fields (although some of the fields from previous MPNN versions may not be included).

We are also looking into the CIF visualization issue. RF3 can take CIF as input; are the MPNN CIF files that you have compatible with RF3 CIF input? If not, please let us know.

MPNN only allows for one of fixed_residues, fixed_chains, designed_residues, or designed_chains to be specified. If you specify one of the design options, the remaining residues should be fixed; if you specify one of the fixed options, the remaining residues should be designed.