Membrane proteins—including GPCRs, ion channels, and transporters—present unique challenges for structure prediction. Learn how to predict, validate, and analyze transmembrane protein structures with AlphaFold2.
#Why Membrane Proteins Matter
Membrane proteins represent:
- ~30% of all proteins in genomes
- ~60% of current drug targets
- <3% of structures in PDB (difficult to crystallize)
Major Classes
- GPCRs: G protein-coupled receptors (drug targets)
- Ion channels: Voltage-gated, ligand-gated
- Transporters: ABC, MFS, solute carriers
- Receptors: Receptor tyrosine kinases
#AlphaFold2 Performance on Membrane Proteins
What Works Well
High Accuracy For
- Transmembrane α-helical bundles (GPCRs, channels)
- β-barrel membrane proteins (porins)
- Proteins with good MSA coverage
- Overall topology and helix packing
Challenges and Limitations
Be Cautious With
- Conformational states (active vs. inactive)
- Lipid-protein interactions
- Ligand-bound states
- Flexible loops between TM helices
#Prediction Workflow
Step 1: Sequence Preparation
Critical considerations for membrane proteins:
- Include signal peptides if not cleaved
- Check for correct isoform (splice variants)
- Identify known TM regions (use TMHMM, Phobius)
bash
# Example GPCR sequence
>ADRB2_HUMAN Beta-2 adrenergic receptor
MGQPGNGSAFLLAPNGSHAPDHDVTQERDEVWVVGMGIVMSLIVLAIVFGNVLVITAIAKFERLQTVTNYFI
TLSLADLVMGLLVVPFGATIVVWGRWEYGSFFCELWTSVDVLCVTASIETLCVIAVDRYFAITSPFKYQSLL
TKNKARVIILMVWIVSGLTSFLPIQMHWYRATHQEAINCYANETCCDFFTNQAYAIASSIVSFYVPLVIMVF
VYSRVFQEAKRQLQKIDKSEGRFHVQNLSQVEQDGRTGHGLRRSSKFCLKEHKALKTLGIIMGTFTLCWLPF
FIVNIVHVIQDNLIRKEVYILLNWIGYVNSGFNPLIYCRSPDFRIAFQELLCLRRSSLKAYGNGYSSNGNTG
EQSGYHVEQEKENKLLCEDLPGTEDFVGHQGTVPSDNIDSQGRNCSTNDSLLStep 2: Topology Prediction
Before running AlphaFold2, predict TM topology:
python
# Using TMHMM via biopython
from Bio import SeqIO
import requests
def predict_tm_topology(sequence):
"""Predict TM helices with TMHMM"""
url = "https://services.healthtech.dtu.dk/cgi-bin/webface2.fcgi"
# Submit to TMHMM server
# Returns TM helix positions
# Example output:
# TM1: residues 31-53
# TM2: residues 63-85
# TM3: residues 103-125
# ...
# TM7: residues 298-320 (for GPCR)Topology Validation
Compare predicted TM regions with AlphaFold2 pLDDT scores. TM helices should have high confidence (pLDDT > 80).
Step 3: Running AlphaFold2
Standard AlphaFold2 settings work well for most membrane proteins. Key points:
- No special membrane-specific settings needed
- AlphaFold2 learned membrane protein features from training data
- Run all 5 models to assess consistency
#Structure Validation
Topology Validation
Verify TM helix predictions match biological expectations:
- Helix length: Typically 20-25 residues for α-helical TM
- Orientation: Hydrophobic residues face lipid bilayer
- Tilt angle: Most TM helices tilt 10-30°
python
from Bio.PDB import PDBParser, calc_dihedral
import numpy as np
def calculate_helix_tilt(structure, helix_start, helix_end):
"""Calculate TM helix tilt relative to membrane normal (z-axis)"""
# Get Cα atoms for helix
ca_atoms = [residue['CA'] for residue in
structure[helix_start:helix_end]]
# Fit line to Cα positions
coords = np.array([atom.coord for atom in ca_atoms])
# Calculate principal axis
# Angle between axis and z-direction is tilt
return tilt_angleMembrane Insertion Analysis
Check if TM regions have appropriate hydrophobicity:
- Use hydrophobicity scales (Wimley-White, Kyte-Doolittle)
- TM regions should have positive hydrophobicity
- Loops should be more hydrophilic
PPM Server
Use the PPM (Positioning of Proteins in Membranes) server to:
- Position predicted structure in lipid bilayer
- Calculate membrane insertion energy
- Identify interfacial residues
#GPCR-Specific Analysis
Conserved Motifs
GPCRs have highly conserved structural motifs:
- DRY motif: D(E)R(K)Y at TM3-IL2 junction
- NPxxY motif: In TM7
- CWxP motif: In TM6
- Disulfide bridge: Between ECL2 and TM3
python
# Check for conserved GPCR motifs
def check_gpcr_motifs(sequence, structure):
"""Validate GPCR-specific structural features"""
# Check DRY motif (end of TM3)
dry_region = sequence[130:133] # Adjust for your protein
assert 'DRY' in dry_region or 'ERY' in dry_region
# Check disulfide bridge
cys1 = find_residue(structure, 'CYS', tm3_position)
cys2 = find_residue(structure, 'CYS', ecl2_position)
distance = calc_distance(cys1['SG'], cys2['SG'])
assert distance < 3.0 # Should form disulfide
# Check NPxxY motif
npxxy_region = sequence[310:315]
assert 'NP' in npxxy_region and 'Y' in npxxy_region[-1]
return TrueConformational State Prediction
Active vs. Inactive States
AlphaFold2 typically predicts inactive/resting state GPCRs:
- Trained mostly on antagonist-bound structures
- Active states require G protein or β-arrestin
- Use AlphaFold2-Multimer with G protein for active states
#Ion Channels
Pore Architecture
For voltage-gated and ligand-gated channels:
- Identify pore-forming regions (usually TM5-TM6 for VGICs)
- Check selectivity filter geometry
- Analyze gate position (open/closed)
- Verify fourfold/fivefold symmetry where expected
Oligomeric Channels
Many channels are homomers or heteromers:
bash
# Tetrameric channel prediction
>subunit_A
MLVPSRLQR...
>subunit_B
MLVPSRLQR...
>subunit_C
MLVPSRLQR...
>subunit_D
MLVPSRLQR...Symmetry Analysis
Use AlphaFold2-Multimer and check:
- ipTM > 0.7 between subunits
- Symmetric arrangement in PAE matrix
- Consistent pore diameter across models
#Transporters
Alternating Access Mechanism
Most transporters alternate between inward- and outward-facing states:
- AlphaFold2 may predict either state
- Check which gates are open/closed
- Substrate binding site accessibility
ABC Transporters
For ABC transporters, predict:
- TMD (transmembrane domain) + NBD (nucleotide-binding domain)
- Check TMD-NBD interface
- Verify NBD dimerization in ATP-bound state
#Lipid Bilayer Context
Molecular Dynamics in Membrane
Refine AlphaFold2 predictions with MD simulations:
bash
# GROMACS workflow for membrane protein
# 1. Insert protein into POPC bilayer
gmx insert-molecules -f protein.pdb -ci POPC.gro -o system.gro
# 2. Add water and ions
gmx solvate -cp system.gro -o solvated.gro
gmx genion -s topol.tpr -o ions.gro
# 3. Energy minimization and equilibration
# 4. Production MD (100-500 ns)
# Analyze:
# - TM helix stability
# - Lipid-protein interactions
# - Conformational samplingMembrane Analysis Tools
- CHARMM-GUI: Membrane builder for MD simulations
- MemProtMD: Database of membrane protein simulations
- PPM: Positioning proteins in membranes
- MEMPROT: Membrane protein topology prediction
#Case Studies
Case 1: β2-Adrenergic Receptor
bash
Target: ADRB2 (GPCR, 413 residues)
AlphaFold2 results:
- pTM: 0.89
- TM regions pLDDT: 92-95 (excellent)
- Loop regions pLDDT: 65-75 (moderate)
Validation:
- Topology matches TMHMM prediction
- RMSD to crystal structure: 1.3 Å (TM bundle)
- Predicted inactive state (correct, no G protein)
- All conserved motifs present and correctCase 2: Voltage-Gated Potassium Channel
bash
Target: Kv1.2 homotetramer
AlphaFold2-Multimer:
- ipTM: 0.82 (high confidence)
- Fourfold symmetry preserved
- Selectivity filter geometry correct
Challenges:
- Voltage sensor in intermediate position
- MD simulation showed transition to activated state
- Required 200 ns simulation for proper pore opening#Drug Discovery Applications
Binding Site Identification
Use predicted structures for:
- Identifying orthosteric binding site
- Discovering allosteric sites
- Structure-based virtual screening
- Designing selective inhibitors
Predict Membrane Protein Structure
Use Protogen Bio's specialized membrane protein tools
#Best Practices Summary
Membrane Protein Checklist
- ✓ Predict TM topology before AlphaFold2 (TMHMM)
- ✓ Validate TM regions have high pLDDT (> 80)
- ✓ Check conserved motifs (GPCRs, channels)
- ✓ Analyze membrane insertion with PPM
- ✓ Use Multimer for oligomeric proteins
- ✓ Consider MD simulations in lipid bilayer
- ✓ Validate conformational state prediction
