Week 4 HW: Protein Design Part I

Part A. Conceptual Questions

Answer any NINE of the following questions from Shuguang Zhang:

  1. How many molecules of amino acids do you take with a piece of 500 grams of meat? (on average an amino acid is ~100 Daltons)
  2. Why do humans eat beef but do not become a cow, eat fish but do not become fish?
  3. Why are there only 20 natural amino acids?

The natural amino acids are determined by codons, which are determined by three nucleotides (of which can be adenine, uracil, guanine, cytosine). This gives 4 x 4 x 4 = 64 total codons, but redundancy among codons produces only 20 unique amino acids.

  1. Can you make other non-natural amino acids? Design some new amino acids.
  2. Where did amino acids come from before enzymes that make them, and before life started?
  3. If you make an α-helix using D-amino acids, what handedness (right or left) would you expect?
  4. Can you discover additional helices in proteins?
  5. Why are most molecular helices right-handed?
  6. Why do β-sheets tend to aggregate?
    • What is the driving force for β-sheet aggregation?
  7. Why do many amyloid diseases form β-sheets?
    • Can you use amyloid β-sheets as materials?
  8. Design a β-sheet motif that forms a well-ordered structure.

Part B: Protein Analysis and Visualization

  1. Briefly describe the protein you selected and why you selected it.

I picked crystallin, which is a protein in the eye responsible for the movement of your iris as you focus. It’s notably transparent, being part of the eye lens, and water-soluble, which was a callback to our lecture. I picked the protein because I was interested in how cataracts were formed.

The specific protein I went with for the following questions is P02511, or Alpha-crystallin B (in humans).

  1. Identify the amino acid sequence of your protein.

The AA sequence from UnitProt is

sp|P02511|CRYAB_HUMAN Alpha-crystallin B chain OS=Homo sapiens OX=9606 GN=CRYAB PE=1 SV=2 MDIAIHHPWIRRPFFPFHSPSRLFDQFFGEHLLESDLFPTSTSLSPFYLRPPSFLRAPSW FDTGLSEMRLEKDRFSVNLDVKHFSPEELKVKVLGDVIEVHGKHEERQDEHGFISREFHR KYRIPADVDPLTITSSLSSDGVLTVNGPRKQVSGPERTIPITREEKPAVTAAPKK

  • How long is it? What is the most frequent amino acid? You can use this Colab notebook to count the frequency of amino acids.

Using the Colab notebook, the protein is 175 amino acids long with the most common amino acid being P (and appearing 17 times).

  • How many protein sequence homologs are there for your protein? Hint: Use Uniprot’s BLAST tool to search for homologs.

According to UniProt, it’s part of the small heat shock protein (HSP20) family, along with all other Alpha-crystallin B proteins. However, according to the Transporter Classification Database, it’s part of the α-Crystallin Chaperone (CryA) family (where other Alpha-crystallin B proteins don’t appear).

  • Does your protein belong to any protein family?

Homology refers to protein sequences that likely have a common ancestor (identified through having similarities in sequence/structure?). Using the BLAST software gives 250 results for similar proteins, with results primarily appearing to be Alpha-crystallin B in different animals.

  1. Identify the structure page of your protein in RCSB

This step was particularly difficult for me, as I didn’t always understand how to get to the answer based on what I had on the screen.

  • When was the structure solved? Is it a good quality structure?

The structure seems to be initially solved in 2009 but has increased members up until 2025. Some particularly high resolution structures were identified in 2012 and 2014 through X-ray diffraction, with a resolution of 1.0 - 1.5 Å.

  • Are there any other molecules in the solved structure apart from protein?

I’m not entirely sure how to identify this…

  • Does your protein belong to any structure classification family?

Using the Structural Classification website, it belongs to the “Alpha crystallin-like” family, further within the “Hsp20 chaperone-like” family.

  1. Open the structure of your protein in any 3D molecule visualization software:

I chose to use PyMol to open my structure, getting the structure below.

  • Visualize the protein as “cartoon”, “ribbon” and “ball and stick”.
Opened structure on PyMol Opened structure on PyMol
  • Color the protein by secondary structure. Does it have more helices or sheets?

The protein seems to mostly be composed of sheets with some helices. Secondary structure colored protein Secondary structure colored protein

  • Color the protein by residue type. What can you tell about the distribution of hydrophobic vs hydrophilic residues?

I colored hydrophobic residues in red and hydrophillic residues in green.

  • Hydrophobic (in red): glycine (Gly), alanine (Ala), valine (Val), leucine (Leu), isoleucine (Ile), proline (Pro), phenylalanine (Phe), methionine (Met), tryptophan (Trp)
    • PyMol code: select hydrophobic, resn Gly resn Ala resn Val resn Leu resn Ile resn Pro resn Phe resn Met resn Trp
  • Hydrophilic (in green): serine (Ser), threonine (Thr), asparagine (Asn), glutamine (Gln), cysteine (Cys), glycine (Gly)
    • PyMol code: select hydrophillic, resn Ser resn Thr resn Asn resn Gln resn Cys resn Gly

Residue-colored protein Residue-colored protein I had to switch to a spheres visualization to better see how molecules were interacting. It was a little hard for me to see a significant pattern, but I do feel like the hydrophilic residues have more “open” facing areas, whereas the hydrophobic residues were more clumped (both together and with neighboring residues).

  • Visualize the surface of the protein. Does it have any “holes” (aka binding pockets)?

Visualizing the protein as a surface was really helpful! I could easily find a couple areas that could be binding pockets. It’s a little difficult to show it accurately in a photo, but I indicated potential areas below:

Surface visualization of protein Surface visualization of protein

Part C. Using ML-Based Protein Design Tools

In this section, we will learn about the capabilities of modern protein AI models and test some of them in your chosen protein.

  1. Copy the HTGAA_ProteinDesign2026.ipynb notebook and set up a colab instance with GPU.
  2. Choose your favorite protein from the PDB.
  3. We will now try multiple things in the three sections below; report each of these results in your homework writeup on your HTGAA website:

I had to pivot here as I originally chose vicilin from lentil beans (Lens culinaris), which was a type of globulin storage protein. It has a rather long sequence (shown below) and as a result made reading some results pretty challenging. I ended up switching back to crystallin from Part B. Attempt at a deep mutational scan with vicilin. I had trouble reading the result. Attempt at a deep mutational scan with vicilin. I had trouble reading the result.

C1. Protein Language Modeling

  1. Deep Mutational Scans
    1. Use ESM2 to generate an unsupervised deep mutational scan of your protein based on language model likelihoods.
    2. Can you explain any particular pattern? (choose a residue and a mutation that stands out)
    3. (Bonus) Find sequences for which we have experimental scans, and compare the prediction of the language model to experiment.

sp|P02511|CRYAB_HUMAN Alpha-crystallin B chain OS=Homo sapiens OX=9606 GN=CRYAB PE=1 SV=2 MDIAIHHPWIRRPFFPFHSPSRLFDQFFGEHLLESDLFPTSTSLSPFYLRPPSFLRAPSW FDTGLSEMRLEKDRFSVNLDVKHFSPEELKVKVLGDVIEVHGKHEERQDEHGFISREFHR KYRIPADVDPLTITSSLSSDGVLTVNGPRKQVSGPERTIPITREEKPAVTAAPKK

Deep mutational scan of crystallin Deep mutational scan of crystallin

Proteins in the latter ⅔ of the sequence, particularly 87 (L), 89 (V), 96 (I), 98 (V), 139 (G), 141 (L), and 143 (V), seem subject to severely detrimental mutations (looking at the dark blue streaks). This indicates to me that Leucine and Valine are rather important residues that should avoid mutation. Meanwhile, the first ⅓ of the sequence seems pretty tolerant to any changes. Notably, protein 129 seems like it can be mutated with mostly beneficial outcomes.

  1. Latent Space Analysis
    1. Use the provided sequence dataset to embed proteins in reduced dimensionality.
    2. Analyze the different formed neighborhoods: do they approximate similar proteins?
    3. Place your protein in the resulting map and explain its position and similarity to its neighbors.

3D t-SNE Visualization (done with Plotly) in the Colab. Embedding visualization of proteins Embedding visualization of proteins Shoutout to Nourelden Rihan for his helpful guide on the forum! I was able to plot my protein pretty easily thanks to him. Embedding visualization of proteins + crystallin Embedding visualization of proteins + crystallin My protein was pretty close to a cluster of other proteins, seen in the photo below. Crystallin + its neighbors Crystallin + its neighbors

C2. Protein Folding

Folding a protein

  1. Fold your protein with ESMFold. Do the predicted coordinates match your original structure?

I would say it looks pretty similar. On the left is the ESMFold result, the middle is a structure derived experimentally (on PDB), and on the right is a structure derived computationally (on PDB). You can see ESMFold has a similar sheet structure with the other two, but the placement of the loops (especially the bottom one) is identical to the right structure (both theoretical) while not necessarily resembling how the structure looks in the middle photo. Embedding visualization of proteins Embedding visualization of proteins

  1. Try changing the sequence, first try some mutations, then large segments. Is your protein structure resilient to mutations?

C3. Protein Generation

Inverse-Folding a protein: Let’s now use the backbone of your chosen PDB to propose sequence candidates via ProteinMPNN

  1. Analyze the predicted sequence probabilities and compare the predicted sequence vs the original one.
  2. Input this sequence into ESMFold and compare the predicted structure to your original.

This part was pretty confusing to me, as the file I’d been referring to earlier (a 175 long sequence for Alpha-crystallin B, called P02511) wasn’t available on PDB. I ended up having to source 2N0K (pdb_00002n0k), a member within P02511 that ultimately had the same sequence, despite other members having different sequences.

I prompted the Colab to design for chain A and B. Both had a length of 82 AA (164 total) and had the following potential amino acid variations. Amino acid probabilities with Chain A in Crystallin Amino acid probabilities with Chain A in Crystallin The new sequence ended up being

PATPEERTIELKVPNAKPENIEVIIDGGRITVKAKELVEKRENCDYYKGYLVECDDPERVDPETMKAEIDEDGTVTIYGPGAPATPEERTIELKVPNAKPENIEVIIDGGRITVKAKELVEKRENCDYYKGYLVECDDPERVDPETMKAEIDEDGTVTIYGPGA

After plugging into ESMFold, I got a structure that looks pretty similar to the original, at least in the way the sheets fold on each other. ESMFold of new sequence ESMFold of new sequence

Part D. Group Brainstorm on Bacteriophage Engineering

Our proposal (and research notes) for this assignment can be accessed here. (WIP)