Week 3
Structure prediction: the classical formulation
Structure prediction is posed as an optimization problem: find the conformation that minimizes an energy or scoring function over an enormous search space. Classical methods differ by how much they can borrow from known structures. Homology modeling applies when a close template exists, threading in the twilight zone, and de novo methods otherwise. This week sets up that formulation and teaches the first tool every template-based method needs: aligning a query sequence to a template by dynamic programming.
Objectives
- State protein structure prediction as an optimization problem by defining its representation (Cartesian or internal coordinates), search space, and objective (energy or score). Explain how Anfinsen's thermodynamic hypothesis justifies it.
- Assign a target to homology modeling, fold recognition, or de novo prediction from its sequence identity to the closest known structure, using the safe, twilight, and midnight zones.
- Score a pairwise alignment with a substitution matrix (PAM or BLOSUM) and linear or affine gap penalties, and explain why similar residues score higher than dissimilar ones.
- Fill the dynamic programming matrix and trace back an optimal global alignment (Needleman–Wunsch), and state its O(mn) time and space cost.
- Modify the recurrence for local alignment (Smith–Waterman) and for affine gaps (three-state Gotoh), and say when local alignment is the right choice.
- Explain why database searches use heuristics such as BLAST and FASTA instead of full dynamic programming, and use a BLAST search against the PDB to pick a modeling template.
- List the eight steps of homology modeling (template recognition, alignment correction, backbone generation, loop modeling, side-chain modeling, optimization, validation, iteration) and say what can go wrong at each.
- Use a multiple sequence alignment to correct a pairwise alignment in low-identity regions and to place insertions and deletions in variable regions rather than in the core.
- Compare knowledge-based, fragment-based, and energy-based loop modeling. Explain how rotamer libraries make side-chain placement tractable, and check a model with force-field energies and normality indices.
Reading
| Gu & Bourne 30 | Venselaar, Krieger, and Vriend walk through homology modeling step by step: BLAST template search, the residue exchange and alignment matrices, loops, rotamers, optimization, and validation. Focus on Step 1, where the best path through the alignment matrix is the dynamic programming problem from lecture 6, and on the identity thresholds in Figure 30.1. |
Lectures
| 5 | Classical formulation of protein structure prediction; optimization transcript | 1:15:05 |
| 6 | Protein sequence alignment and dynamic programming transcript | 1:09:12 |
| 2:24:17 |