Example 3: Compare Crystal and AlphaFold TP53 Structures

The objective of this exercise is to compare a TP53 AlphaFold-predicted structure with experimentally determined structure. By searching for an AlphaFold-predicted structure and understanding the limitations of predicted models, participants will assess model accuracy, identify potential discrepancies, and gain a more comprehensive understanding of TP53.

iCn3D Help Docs

Background on Structure Prediction

The field of structure prediction leverages computational algorithms to generate models of biomolecular structures when experimental data is insufficient or challenging to obtain. It is growing in popularity and accuracy as computational algorithms and facilities improve, presenting new opportunities to combine computational and experimental structure prediction approaches to improve our understanding of biomolecular structure.

AlphaFold in iCn3D

AlphaFold, a machine learning model for protein structure prediction, is now integrated into iCn3D, offering easy access and analysis of AlphaFold predicted structures. While AlphaFold represents a considerable advancement in protein structure prediction,  it's important to consider limitations, such as: 

• Only predicts monomers (AlphaFold Multimer model is published separately as open-source)
• Unreliable for IDP, it does convey the information via a low confidence score
• Not validated for mutational analysis
• Can only output one conformation of proteins with multiple conformations
• Only predicts the main peptide chain, not missing co-factors, metals, and co- and post-translational modifications

pLDDT: Confidence Scores in AlphaFold Structures

AlphaFold assigns a pLDDT score (0-100) to each residue, indicating confidence in its predicted structure. This score is stored in the B-factor field of PDB files. The confidence levels are:

• High Accuracy (90+): Suitable for atomistic experiments (highly reliable)
• Good (70-90): Generally reliable, especially for backbone structure
• Low Confidence (50-70): Use with caution
• Low (Below 50): Unreliable, avoid interpretation

Compare TP53 Crystal (PDB 1TUP) and AlphaFold Structure

1. Go to iCn3D and load PDB structure "1TUP"

2. Go to File > Search Structure > AlphaFold Structures.

3. Input Protein/Gene Name "TP53: CELLULAR TUMOR ANTIGEN P53" and select Search. An AlphaFold structure should load.

• iCn3D searches the RefSeq database using the input (Protein/Gene Name) and prioritizes search results where the corresponding RefSeq protein has a matching AlphaFold UniProt ID.

• You can view information about the structure on the AlphaFold Data base by selecting the underlined title: TP53: CELLULAR TUMOR ANTIGEN P53

Note: Due to ongoing database updates and the potential of multiple models available per protein, the specific AlphaFold accession (e.g., E3U906) you see may differ

4. Render the structure in a meaningful way.

   Hint: Change the Style and/or Color to your liking and review Sequence and Annotations Details.
   Follow
   Answer the following questions:

   • How does this structure differ from the crystal structure (PDB 1TUP)?

Hint: what biomolecules are present in the crystal structure that are not present in the AlphaFold structure?

• TP53 is known to form functional dimers and tetramers to help with high-affinity DNA binding, but AlphaFold only shows a monomer. How can this limitation influence your understanding of TP53 function?

5. Review the color by pLDDT legend.

   Answer the following question:

   • Which residues have high and low pLDDT? Why might this be?

     Hint: Loop regions are generally more challenging to resolve in crystal structures and predict accurately via structure prediction methods. Also, the alignment window “Select residues in aligned sequences” will show primary structure (amino acid codes) colored by pLDDT. You can use this window and the 3D rendering to help answer the question.
     Follow

6. To align the predicted and crystal structures select File > Align > Multiple Chains > By Structure. In Chain IDs input "1TUP_B,E3U906" (PDB Chain,AlphaFold Accession) and select Align with TM-align.

   Answer the following question:

   • What do the Alignment RMSD and TM-score indicate about these structures?

Hint: TM-align compares the overall similarity and calculates the structural alignment between two protein structures (crystal structure and AlphaFold model).

• Alignment RMSD is a calculation of the average distance between equivalent Cα atoms in the aligned protein structures, reflecting overall backbone similarity. <2 Å generally indicates a high degree of similarity in the backbone conformation.

• TM-score is a measure of the similarity between the overall folds of the aligned protein structures. >0.7 indicates high overall fold similarity, 0.5-0.7 suggests partial similarity with potential regional deviations, and below 0.5 suggests low structural similarity.

7. Change Style and Color to your liking.

Hint: Color the crystal structure (1TUP) in a solid, neutral color and your AlphaFold structure (e.g., E3U906) by pLDDT
Follow

8. Participants working on the other exercises have learned 1) residues P177, H178, H179, and R181 are important for dimerization and 2) mutations at K120 (K>A, K>E, K>Q) can potentially disrupt DNA binding, indicating K120’s importance in TP53 function. Review these residues (P177, H178, H179, R181, and/or K120) in crystal and AlphaFold structures. Focus on as many as time allows.

   Answer the following question:

   • Are there any significant differences observed between the crystal and AlphaFold structures for these residues?

9. Render an informative visualization (e.g., highlighting a key difference or agreement) to illustrate your observations.