Protein Chemistry

day 11 alphafold

🧬 AlphaFold — Fun & Educational Summary (Theory Only)

This file is essentially a conceptual introduction to AlphaFold, focusing on how to interpret its outputs, not how to run it. Below is a structured, detailed explanation of all key topics—including the ones you explicitly asked about and the ones you didn’t.

🧠 1. What is AlphaFold?

Core idea

AlphaFold predicts 3D protein structures from amino acid sequences.
Input: protein sequence
Output: predicted 3D structure

Why it works so well

Uses huge datasets (e.g., PDB structures) + machine learning
Key breakthrough: massive computing power + learning patterns from known structures

Important distinction

There are two ways AlphaFold is used:

Database/server → already predicted structures (you just look them up)
Prediction tool → input your sequence and generate a model

📊 2. plDDT Score (VERY IMPORTANT)

What is plDDT?

Per-residue confidence score
Measures: 👉 How reliable is the predicted local structure at each amino acid?

🔍 How to interpret plDDT

Color	Meaning	Interpretation
🔵 Blue	High confidence	Very reliable structure
🟡 Yellow	Medium confidence	Uncertain
🔴 Red	Low confidence	Likely unreliable / flexible

Your specific questions answered

❓ “plDDT score → how reliable is the structure?”

✔ Yes — but only locally, not globally

❓ “plDDT high → alpha helices?”

Not exactly causal, but:
- Alpha helices are stable and well-defined
- → often predicted with high plDDT
So: ✔ High plDDT often corresponds to structured regions (like helices)

❓ “plDDT high = less certain?”

❌ Incorrect ✔ It is the opposite:

High plDDT → high confidence
Low plDDT → low confidence

❓ “plDDT → local structure?”

✔ Correct

It evaluates each residue independently
Does NOT tell you if the overall fold is correct

📈 plDDT plot

X-axis: residue number
Y-axis: confidence score
Used to:
- Identify stable vs flexible regions
- Predict disordered regions

🧠 Key interpretation insight

Yellow/red regions are often:
- Flexible loops
- Disordered regions
- Not necessarily “wrong”—just not well-defined structurally

🎨 3. Confidence-Colored Structure

This is just a visual representation of plDDT:

Structure is colored:
- Blue = reliable
- Yellow/red = uncertain

👉 Helps you quickly see which parts to trust

🌍 4. PAE Heatmap (Global Confidence)

What is PAE?

Predicted Aligned Error

Measures: 👉 How confident is AlphaFold about the relative positions between residues?

🔍 Key idea

plDDT = local confidence
PAE = global confidence

📊 How to read a PAE heatmap

Axes: residue vs residue
Color:
- Dark (low error) → high confidence
- Bright (high error) → low confidence

❓ “PAE → global structure?”

✔ Yes

It tells:

Whether different parts of the protein are positioned correctly relative to each other

❓ “PAE → domain placement?”

✔ Exactly

Especially important for multi-domain proteins

🧩 5. Multi-domain Proteins & AlphaFold Limitations

❓ “Can AlphaFold place multiple domains correctly?”

👉 Answer: Sometimes—but not always

What happens:

Each domain: ✔ Often predicted correctly (high plDDT)
BUT: ❗ Relative positioning between domains:
- May be uncertain (high PAE)

🧠 Interpretation

AlphaFold might show:
- Domain A here
- Domain B here

BUT:

Real structure could be:
- Domain B rotated, shifted, or interacting differently

🔥 Key takeaway

Trust domains individually
Be cautious about how domains are arranged

⚠️ 6. Important Limitations

From the lecture:

1. Complexes (older versions)

Early AlphaFold struggled with:
- Protein–protein interactions
Improved in newer versions (e.g., AlphaFold 3)

2. Flexible regions

Low plDDT → not necessarily wrong
Often:
- Intrinsically disordered regions

3. Global uncertainty

Even if:
- Local structure = perfect
The overall fold may still be uncertain

🧪 7. What You Should Know for Exam

From the lecture emphasis:

You need to know:

What AlphaFold does
plDDT:
- Meaning
- Local vs global
- Color interpretation
PAE:
- What it measures
- Domain positioning
Difference:
- Local vs global confidence

You do NOT need:

Deep algorithm details
How to run AlphaFold
Advanced features

🧠 Final Conceptual Summary

AlphaFold workflow (simplified)

Input sequence
AI predicts structure
Outputs:
- 🧬 3D model
- 📊 plDDT (local confidence)
- 🌍 PAE (global confidence)

🚨 Golden Rules

🔵 High plDDT → trust that region
🔴 Low plDDT → flexible or uncertain
🟢 Low PAE → domains positioned reliably
🔴 High PAE → domain arrangement uncertain

🧩 Mental model

Think of AlphaFold like this:

plDDT = “Are the LEGO bricks correct?”
PAE = “Did we assemble the LEGO pieces correctly?”

🧬 Core AlphaFold Knowledge (Exam-Focused)

1. What AlphaFold does

Predicts a 3D protein structure from an amino acid sequence
Uses patterns learned from known structures (e.g., PDB)
Output is not just a structure, but also confidence metrics

👉 Key idea: It gives you a model + information about how much you should trust it

2. plDDT — Local Confidence Score

Meaning

plDDT = per-residue confidence score
Indicates how reliable the structure is at each amino acid

Local vs Global

plDDT is local
- Evaluates each residue independently
- Does NOT tell you if the whole protein is assembled correctly

Color Interpretation

Color	Confidence	Meaning
🔵 Blue	High	Very reliable structure
🟡 Yellow	Medium	Some uncertainty
🔴 Red	Low	Unreliable / flexible region

Interpretation Rules

High plDDT:
- Structure likely correct
- Often corresponds to ordered regions (e.g., helices, sheets)
Low plDDT:
- Region may be:
  - Flexible
  - Disordered
  - Not adopting a fixed structure

Important Insight

Low plDDT ≠ wrong
It often means the protein itself is flexible, not that AlphaFold failed

3. PAE — Global Confidence

Meaning

PAE = Predicted Aligned Error
Measures how well different residues are positioned relative to each other

What it Measures

👉 Not structure itself, but:

Confidence in distances between residues

Local vs Global

PAE is global
- Evaluates relationships between regions/domains
- Tells if the overall fold is reliable

Domain Positioning

Especially important for multi-domain proteins

Example:

Domain A → high plDDT → correct structure
Domain B → high plDDT → correct structure

BUT:

PAE high between A and B → ❗ Their relative placement is uncertain

Interpretation

PAE Value	Meaning
Low	Domains positioned correctly
High	Domains may be misplaced

4. Difference: plDDT vs PAE

Feature	plDDT	PAE
Type	Local	Global
Measures	Residue confidence	Relative positioning
Use	Trust specific regions	Trust overall fold
Key limitation	Ignores domain arrangement	Doesn’t show local accuracy

🧠 Simple Mental Model

plDDT → “Is each piece built correctly?”
PAE → “Are the pieces assembled correctly?”

5. Why α-Helix is More Stable than β-Sheet (and others)

This is where your reasoning needs to be precise—there’s a common oversimplification here.

❗ First correction

It is not universally true that α-helices are always more stable than β-sheets.

Stability depends on:

Environment (solution vs membrane)
Sequence
Interactions

However, in many contexts (especially isolated peptides), α-helices appear more stable/predictable.

🔬 Reasons α-Helices Often Appear More Stable

1. Local Hydrogen Bonding

α-helix:
- H-bonds form between i → i+4 residues
- → very local and self-contained
β-sheet:
- H-bonds form between different strands
- → requires long-range alignment

👉 Consequence:

Helices fold easily and independently
Sheets require precise positioning

2. Folding Kinetics

α-helix:
- Forms quickly
- Doesn’t depend on distant residues
β-sheet:
- Slower formation
- Needs correct strand pairing

👉 More opportunities for error → less predictable

3. Entropic Cost

β-sheets:
- Require aligning separate chain segments
- → higher entropy loss
α-helices:
- Form within a continuous chain
- → lower entropy penalty

4. Exposure & Stability

β-sheets:
- Often expose backbone → can be less stable alone
- Stabilized by:
  - Packing into cores
  - Interactions with other sheets
α-helices:
- Can be stable even in isolation

🧠 Why this matters for AlphaFold

α-helices:
- Regular geometry
- Strong local patterns
- → easier to predict → often high plDDT
β-sheets:
- Depend on long-range contacts
- → harder to predict → sometimes lower confidence

🧾 Final Summary

What to remember

AlphaFold: predicts structure from sequence
plDDT:
- Local confidence
- Blue = reliable
PAE:
- Global confidence
- Critical for domain arrangement
Difference:
- plDDT = local accuracy
- PAE = global correctness

Structural insight

α-helices:
- Local, self-stabilizing → easier to predict
β-sheets:
- Long-range, cooperative → harder to predict

Quiz

Score: 0/31 (0%)

Q0. What is the main purpose of AlphaFold in the context of the lecture?

To experimentally determine protein structures using X-ray crystallography

To predict protein 3D structures from amino acid sequences

To measure protein concentration by absorbance

To identify protein bands on SDS-PAGE gels

Q1. According to the lecture, what was one major reason AlphaFold became much more successful than earlier protein-structure prediction attempts?

It replaced amino acid sequences with DNA sequences

It ignored known structures and predicted everything randomly

It combined large structural databases with powerful computational algorithms

It only predicted proteins that had already been crystallized

Q2. Which statement best describes the AlphaFold database/server mentioned in the lecture?

It is only used to perform SDS-PAGE analysis

It contains many precomputed predicted protein structures

It only stores experimentally solved structures

It is used only for DNA sequencing

Q3. What does the plDDT score mainly tell you?

How reliable each local residue-level region of the predicted structure is

How strongly two proteins bind to each other

How many domains the protein contains

Whether the protein was purified correctly

Q4. Why is plDDT considered a local confidence measure?

Because it measures the total molecular weight of the protein

Because it evaluates confidence residue by residue rather than the overall arrangement of domains

Because it only works for membrane proteins

Because it compares the protein to DNA sequence databases

Q5. In AlphaFold confidence coloring, what does a blue region usually indicate?

Low confidence and likely disorder

High confidence in the local predicted structure

A region that must be experimentally wrong

A region that cannot form secondary structure

Q6. What does a yellow or red region in a confidence-colored AlphaFold structure often suggest?

The protein is definitely crystallized

The region is always an alpha helix

The region is less certain and may be flexible or disordered

The protein has no amino acids in that region

Q7. A student says: 'High plDDT means AlphaFold is less certain about the structure.' What is the correct response?

Correct; high plDDT means low confidence

Incorrect; high plDDT means high local confidence

Correct only for alpha helices

Correct only for multi-domain proteins

Q8. Why might alpha helices often appear with high plDDT scores in AlphaFold predictions?

Because alpha helices are regular, locally stabilized structures that are often easier to predict

Because AlphaFold automatically labels all blue regions as helices

Because helices never occur in flexible regions

Because beta sheets cannot have high plDDT

Q9. What is the main purpose of a plDDT plot?

To show confidence values along the protein sequence residue by residue

To show only the protein's molecular mass

To show the number of genes in the organism

To show how fast the protein was purified

Q10. What does PAE stand for in the AlphaFold output?

Protein Alignment Energy

Predicted Amino-acid Error

Predicted Aligned Error

Protein Activity Estimate

Q11. What does the PAE heatmap mainly help evaluate?

Whether the protein was digested by trypsin

Whether different regions or domains are positioned correctly relative to each other

Whether individual residues are colored blue or red by plDDT

Whether the protein sequence contains stop codons

Q12. Why is PAE especially important for multi-domain proteins?

Because multi-domain proteins cannot have plDDT scores

Because each domain may be predicted well locally, but their relative arrangement may still be uncertain

Because PAE measures protein concentration

Because PAE replaces the amino acid sequence

Q13. Which comparison best describes the difference between plDDT and PAE?

plDDT measures local confidence; PAE measures confidence in relative/global positioning

plDDT measures DNA quality; PAE measures RNA quality

plDDT is experimental; PAE is theoretical

plDDT is only for beta sheets; PAE is only for alpha helices

Q14. If two domains both have high plDDT but high PAE between them, what is the best interpretation?

Both domains are probably locally reliable, but their relative placement is uncertain

The whole structure is certainly correct

The protein has no flexible regions

The sequence must be wrong

Q15. AlphaFold predicts protein 3D structure from amino acid sequence.

True

False

Q16. The lecture mainly emphasized that students must understand the detailed programming architecture of AlphaFold for the exam.

True

False

Q17. The AlphaFold database can contain already predicted structures, meaning the user may not always be generating a new prediction from scratch.

True

False

Q18. plDDT tells you whether the global arrangement of multiple domains is definitely correct.

True

False

Q19. A high plDDT score means AlphaFold has high confidence in the local structure of that region.

True

False

Q20. A low plDDT score always means the amino acid sequence is incorrect.

True

False

Q21. Blue coloring in an AlphaFold confidence-colored structure usually indicates high local confidence.

True

False

Q22. Yellow and red regions in AlphaFold confidence coloring are usually interpreted as lower-confidence regions.

True

False

Q23. A plDDT plot can help identify which parts of the sequence are predicted with high or low confidence.

True

False

Q24. PAE is mainly used to assess local residue-level confidence in secondary structure formation.

True

False

Q25. Low PAE between two regions means AlphaFold is more confident about their relative placement.

True

False

Q26. High PAE between domains can mean that AlphaFold is uncertain about how those domains are arranged in the full structure.

True

False

Q27. For a multi-domain protein, AlphaFold may predict each domain well but still be uncertain about the overall domain arrangement.

True

False

Q28. The older/basic AlphaFold versions had limitations when predicting protein complexes.

True

False

Q29. AlphaFold predictions should always be interpreted together with their confidence scores rather than treated as unquestionable experimental structures.

True

False

Q30. The most important exam-relevant AlphaFold concepts from the lecture are basic understanding of AlphaFold, plDDT, PAE, local confidence, global confidence, and domain positioning.

True

False