A wet-lab researcher looks at an MD model and says, 'Nice cartoon, now show me the data from a universe that actually exists.
Modeling Tau-F sticking core; Algorithm workflow.
1.
Input.
-
The Tau8 isoform (Tau-F, P10636-8) sticking core is known as a sequence of 95 amino acids.
Tau8-95 = "IKHVPGGGSVQIVYKPVDLSKVTSKCGSLGNIHHKPGGGQVEVKSEKLDFKDRVQSKIGSLDNITHVPGGGNKKIETHKLTFRENAKAKTDHGAE"
2.
Handle shape of intrinsically disordered protein (IDP).
-
AlphaFold Server: The AlphaFold server cannot be used to define the 3D shape of
IDP because its core architecture is fundamentally designed to predict a single,
rigid, static conformation rather than a dynamic ensemble. When AlphaFold is forced
to process an IDP sequence, it typically outputs a chaotic, "ribbon-like" chain of
long extended loops. In newer iterations like AlphaFold 3 (which powers the AlphaFold
Server), the model can occasionally exhibit structural hallucinations. It will try
to "fix" or satisfy the unstructured regions by falsely introducing spurious alpha
helices or elements of secondary structure that do not actually exist in nature.
-
The Reality of IDPs: To truly define an IDPs "structure,"
someone needs
a conformational ensemble as a collection of (ideally: hundreds or thousands)
of structural variations that capture the molecule's spatial fluidity. Forcing
an IDP into a single model lacks physical meaning and often artificially overestimates
structural metrics like the radius of gyration.
-
Adopted Methodology: To generate different pdb-files from the same FASTA sequence,
the algorithm relies on random sampling to change the spatial direction of
each subsequent amino acid backbone atom. Specifically, it changes the structure
from file to file using two core components Random Initial Vector and
Cone-Constrained Random Steps:
-
Random Initial Vector: At the start of each new file, the second
C-alpha atom (Cα) is projected from the origin in a completely
randomized direction using random angles (phi_init and costheta_init).
-
Cone-Constrained Random Steps: For every subsequent amino acid in the
chain, the script randomly generates a new directional step inside
a forward-facing cone (phi and costheta). Because it uses Python's
random.uniform() without a fixed seed, these mathematical choices are
unique for every model.
-
Collision Avoidance: If a random step causes an amino acid to clip or
get too close to an older part of the chain (< 4.0 Å), the algorithm
rejects that step, rewinds a few paces, and randomly guesses a different
forward direction. This 3D shape construction algorithm almost guarantees
that while every single file contains the exact same 95-residue structural
sequence layout, every .pdb-file describes a completely unique, un-knotted
geometric shape.
Note: In this instance, the algorithm was employed to produce a collection
of 25 geometric configurations derived from Tau8-95. A particular configuration
(designated as #22) encountered a failure during a subsequent procedural stage,
ostensibly attributable to a collision event. A possible resolution may involve
modifying the cutoff parameter from 4.0 Å to 5.0 Å.
-
Chemically Valid: The generated structural models strictly adhere to
the fundamental physical and spatial constraints of a real protein backbone.
Specifically, the algorithm prevents two physical impossibilities from happening
in the simulation.
-
Every adjacent Alpha-Carbon (Ca) atom is placed exactly 3.8
Angstroms apart. This perfectly mimics the constant, real-world
distance between alpha-carbons in a peptide chain.
-
There are no steric collisions (self-avoiding Walk). The algorithm
enforces a 4.0 Angstrom minimum distance threshold (5.0 at
improved version). This ensures that as the protein chain randomly
bends and moves, it never loops back and passes directly through
itself. In physics, this prevents two atoms from occupying the same
space at the same time (steric clash).
3.
Add Missing Atoms: Inserts missing heavy atoms by PDBFixer software tool used to repair, clean, and prepare Protein .pdb-files so they can be used cleanly in molecular dynamics (MD) simulations and other structural biology workflows.
4.
Energy minimization and an equilibration: GROMACS was used for getting the 3D shape
of IDP from raw starting structure to a stabilized, ready-to-run state.
-
Energy Minimization:
-
Integrator: steep (Steepest Descent minimization).
-
Purpose: It relaxes the starting coordinates, eliminates steric clashes
between atoms (like protein atoms and water), and brings the system to
a local potential energy minimum before any further simulations.
-
Equilibration:
-
Integrator: md (Molecular Dynamics).
-
Ensemble: NVT (Constant Number of particles, Volume, and Temperature).
-
Key Indicators:
-
define = -DPOSRES, which applies position restraints to hold the protein
backbone steady while the solvent (water and counterions) settles.
-
tcoupl = V-rescale (active temperature coupling) to heat and stabilize
the system at 300 K.
-
Brief 100-picosecond trajectory (50000 steps x 0.002 ps) designed
explicitly to equilibrate the environment.
-
Modern, user-contributed Amber a99SB-disp force field (https://github.com/paulrobustelli/Force-Fields) tailored specifically for intrinsically disordered proteins was used here. It comes pre-bundled with its own fully integrated optimized 4-point water models.
Radius Gyration
The Radius Gyration (R
g) plot shows a wide structural distribution typical of an intrinsically
disordered protein (IDP) ensemble, ranging from highly compact to highly extended states.
While 24 conformations are a small snapshot compared to the thousands typically needed for
a fully converged IDP ensemble, this initial batch gives a clear look at the structural
landscape.
This particular plot Shows:
-
Wide Structural Variance: The individual folds span an Rg from roughly 1.9 nm to 3.8 nm. For an IDP, this vast range is expected and normal. It indicates that the generation method is properly sampling different states rather than getting stuck in a single rigid conformation.
-
The Two Extremes:
-
Folds near 1.9 nm represent highly collapsed, compact globular-like states.
-
Folds near 3.8 nm represent highly extended, random-coil-like states.
-
Ensemble Central Tendency: The red dashed line shows your mean Rg sits at 2.58 nm. The light pink band shows where the majority of your structures fall (+/- 1 Standard Deviation).
Tau8 isoform (Tau-F, P10636-8), 95 amino acids sticking core. 25 Models (#24 failed).
Note: We recommend
ChimeraX
for Model visualization.