🧬

RNA Folding

Applying Discrete Optimization and Rollout to Computational Biology.

🔬

1. The RNA Folding Problem

Given a sequence of nucleotides, we want to "fold" it by introducing pairings that result in a stable structure (minimizing free energy).

Constraints

A common constraint is that pairings should not "cross".

If \( (i_1, i_2) \) is a pair, there should be no pair \( (i_3, i_4) \) such that \( i_1 < i_3 < i_2 < i_4 \).

RNA Structure Representation

graph LR N1((A)) --- N2((U)) N2 --- N3((G)) N3 --- N4((C)) N4 --- N5((A)) N5 --- N6((U)) N1 -.-> N6 N2 -.-> N5 style N1 fill:#fce7f3,stroke:#ec4899 style N6 fill:#fce7f3,stroke:#ec4899 style N2 fill:#e0e7ff,stroke:#6366f1 style N5 fill:#e0e7ff,stroke:#6366f1 linkStyle 5,6 stroke:#ec4899,stroke-width:2px,stroke-dasharray: 5 5

Dashed lines represent pairings (Hydrogen bonds).

⚙️

2. Optimization Formulation

We can model this as a sequential decision problem. At each nucleotide, we have at most 3 choices:

🔓

Open

Start a new pairing.

🔒

Close

Complete an existing pairing.

➡️

Do Nothing

Leave unpaired.

Rollout Application

  • State: Partial folding of the sequence.
  • Base Heuristic: Existing software that generates a "reasonable" complete folding from a partial one.
  • Expert Software: Evaluates the quality (energy) of complete foldings.
🔭

3. Multistep Lookahead

Since the branching factor is small (at most 3), this problem is ideal for multistep lookahead.

Complexity

With \( \ell \)-step lookahead, the number of Q-factors to compute is at most \( 3^\ell \).

This allows for deeper exploration of the folding space compared to simple one-step lookahead.

📺

4. Learn More

AlphaFold Explained (8 min)

Watch Video →

Protein Folding (2 min)

Watch Video →
📝

5. Test Your Knowledge

1. What is the goal of RNA folding?

2. What is a common constraint in RNA folding?

3. How many control choices are typically available at each step?

4. Why is multistep lookahead suitable for this problem?

5. What plays the role of the cost function in the rollout formulation?

Previous

Lecture 24

Next

Lecture 26