Recitation 9: Protein Structure Prediction

We will model the binding of nanobodies to the covid spike protein.

Basic (✓):

• Download the FASTA files for 8CYJ. Although the crystal does not resolve the full spike protein (just the RBD), the sequence is for the full length protein.
• Use the Alphafold3 server to predict the structure of the receptor binding domain (RBD) to one of the nanobodies (your choice, but note which chain you choose).
• Compute how different the nanobody pose predicted by AF3 is from the crystal pose by aligning the top-ranked predicted structure to only the RBD and calculating the RMSD of the nanobody to the true structure (8CYJ) without aligning them.
  • We can use pymol for this, but by default align will compute the minimum RMSD (after alignment) and will run multiple cycles of outlier removal (to generate the visually most attractive alignment).
  • For calculating the RMSD of two selections without aligning them:

    • create an align object with the align command without transforming the object
      align (sele1), (sele2), cycles=0, transform=0, object=aln_obj

    • Run rms_cur using the alignment object\

      rms_cur sele1 & aln_obj, sele2 & aln_obj, matchmaker=-1
      

• • • Note that if your selections have the same atoms in them, you don't need to the aln_obj (which specifies which atom matches to which between the selections).  Setting matchmaker to -1 means rms_cur assume the order atoms are in the selections is how they should be matched up.
• Align the full spike structure generated by AF3 to 6VXX and report its RMSD. Note you will have predicted a single monomer and the actual structure is a trimer.
• Upload clearly labeled images of both cases (RBD aligned w/nanobody, full spike protein aligned with 6VXX) to GradeScope. Include the chain you chose, both calculated RMSDs, and a 1-2 sentence caption describing how close the predictions are to the crystal structures.

Extra (✓+):

• Truncate the spike protein to include only the RBD
• Redo RMSD analysis of nanobodies (as done above) with both AF3 and Boltz (install with pip install boltz[cuda] if you haven't already) using only the RBD sequence instead of the full length sequence.
• In PyMOL, load 8CYJ and your AF3 and Boltz RBD-only predictions, and align each prediction onto 8CYJ using only the RBD. Color the three structures differently and label them. The RBDs will overlap and the nanobodies will show how far each prediction is from the crystal pose.
• Upload this one overlay figure to GradeScope. Include both nanobody RMSDs (unaligned, same chain as above) and a 1-2 sentence caption saying which model is closer to the crystal pose.