# Protein Structure Prediction
protein motion movies and similar datasets have more than pure visual data
- energy kinetics
- cooperative binding
## Prediction Techniques
| Name | Model Type | Description |
| --------- | ----------------- | ------------------------------------------------------------- |
| ESMFold | LLM | Can be run using a single input sequence |
| AlphaFold | Homology Modeling | Relies on multiple sequence alignment and template structures |
### Homology Modeling
1. Model the framework regions
2. Model the CDR loops
3. Model the CDR-H3 loop
- Potentially uses different approaches from steps 1 & 2
## Determining Quality of Prediction
![[Pasted image 20240609074506.png]]
Resolution
Å
experimental
- crystallography not always under physio conditions
computational
- systematic/methodological errors
---
![[Pasted image 20240527225151.png|500]]
**Protein Homology Modeling**
- Do we have any SSM libraries with a closely homologous antibody that we have the structural information of?
![[Pasted image 20240601034610.png|500]]
**From Sequence to 3D Structure**
- PDB structures are only models of a *snapshot*
- technically newer techniques like cryo-em try to circumvent this right?
> [Accelerating Antibody Drug Discovery Through Computational Modeling (Youtube)](https://www.youtube.com/watch?v=4pMb0EJwnFg)
## Post-Prediction Analyses
### Protein Property Analysis
**Scalar Based Properties:** Calculated on the entire structure
- Ex: pI, net charge, hydrodynamic radius, dipole moment, hydrophobic moment, etc.
**Sequence Based Properties**
**Region-Based Properties:** Residue aware properties
- Ex: Charge, Hydrogen Bond Donors/Acceptors, Surface Exposure, Energetic Contributions to Different Patches