lamm-mit/keratin-mech-seq-ss-md-properties

Human Keratin Mechanical, Sequence, SS, and MD Properties

Long-form human keratin mechanical-property and molecular-dynamics results linked to UniProt/AlphaFold keratin records. Each protein has four velocity-specific rows (v0, v1, v2, and v4).

Keratin Protein Background and Study Significance

Keratins are structural proteins that form intermediate filaments in epithelial tissues, including hair, skin, and nails. In hair fibers, keratin proteins assemble into hierarchical structures that help determine mechanical properties such as strength, elasticity, toughness, and resistance to fracture. Keratins are commonly grouped into two complementary classes: Type I keratins, which are generally acidic, and Type II keratins, which are generally basic or neutral. These proteins pair to form coiled-coil heterodimers that further assemble into larger filament networks.

This work focuses on human hair keratins at the molecular scale. While hair-fiber mechanics are often studied experimentally at the macroscopic level, the molecular unfolding behavior of individual keratin proteins is less systematically characterized. The paper addresses this gap by curating 51 human keratin proteins, predicting or collecting their structures, and simulating their unfolding under controlled steered molecular dynamics conditions.

flowchart LR
    A["Human keratin sequences"] --> B["AlphaFold protein structures"]
    B --> C["Equilibration MD"]
    C --> D["Steered MD pulling"]
    D --> E["Force-displacement curves"]
    E --> F["Strength and toughness"]
    B --> G["Sequence and structure descriptors"]
    G --> H["Structure-property analysis"]
    F --> H
    H --> I["Molecular insight into hair-fiber mechanics"]

The significance of this study is that it provides a reproducible, dataset-scale molecular dynamics framework for comparing keratin unfolding mechanics across Type I and Type II keratins. By connecting sequence features, predicted structures, secondary-structure descriptors, and MD-derived mechanical properties, the study helps explain how molecular-scale protein behavior may contribute to the toughness and resilience of hierarchical hair fibers.

The steered molecular dynamics pulling velocities used in the study are accelerated computational probes, not direct reproductions of experimental hair-fiber strain rates. Their value is comparative: they reveal relative unfolding trends, rate-sensitive stiffening, strength-toughness coupling, and relationships between molecular descriptors such as sequence length, molecular weight, coil content, SASA, and energy absorption.

Dataset Summary

• Hub repo: lamm-mit/keratin-mech-seq-ss-md-properties

• Records: 204

• Organism: Homo sapiens (NCBI:txid9606)

• Keratin type counts: {"type_I": 100, "type_II": 104}

• Sequence length range: 394-644 amino acids

• Median sequence length: 493.0

• Unique proteins: 51

• Velocity row counts: {"v0": 51, "v1": 51, "v2": 51, "v4": 51}

• Strength range: 749.59-1799.58

• Toughness range: 449074.0737432281-2086165.5413953725

• Sequence-string discrepancies flagged in records: ["KRT17", "KRT26"]

• Sequence length-field/string discrepancies flagged in records: ["KRT17", "KRT26"]

• Generated at: 2026-06-05T15:18:17.881347+00:00

Fields

• result_id and KRT record_id
• velocity condition
• protein sequence and source-match flags
• force vector, raw force-vector text, checksum, and vector statistics
• strength, toughness, normalized strength, and normalized toughness
• molecular weight, instability index, isoelectric point, SASA
• amino-acid counts/percentages and secondary-structure composition
• crystallinity, secondary-structure transitions, hydrogen-bond change, and persistence length
• record provenance and checksum fields

Dataset Generation

This long-form dataset reports molecular descriptors and nanomechanical outputs from the comparative keratin unfolding workflow described in the accompanying manuscript, Comparative Molecular Dynamics Characterization of Hair Keratin Unfolding Mechanics. Starting from 51 curated human keratin monomer sequences, sequence descriptors were computed with ProtParam-style calculations, including sequence length, molecular weight, instability index, and isoelectric point. Secondary-structure and structural descriptors were computed from folded structures and simulation trajectories, including SASA, crystallinity, secondary-structure composition, transition counts, hydrogen-bond change, and persistence length.

Implicit atomistic molecular dynamics simulations were performed with NAMD using the CHARMM force field and the generalized Born implicit solvent (GBIS) model. Each protein was minimized and equilibrated before steered molecular dynamics. In SMD, one terminus was fixed while tensile pulling was applied to the other terminus until the estimated contour length was reached, using four accelerated pulling velocities: 0.1, 0.125, 0.15, and 0.175 A/ps. These velocities are computational probes for comparing relative molecular-scale unfolding behavior and should not be interpreted as direct reproductions of experimental hair-fiber strain rates.

For each simulation run, the force-displacement response was processed to compute strength as the maximum unfolding force and toughness as the area under the force-displacement curve. The force trace was also resampled to the protein sequence length by segment averaging, producing a per-residue-length force vector suitable for model training or comparative analysis. Normalized strength and toughness values are included for the combined 204-run dataset.

Sources and Attribution

Sequence records are linked to UniProtKB, and structure records are linked to AlphaFold Database entries where available. UniProt states that copyrightable parts of its databases are available under CC BY 4.0, and AlphaFold Database states that its data is available under CC BY 4.0 for academic and commercial use. Users should cite UniProt, AlphaFold Database, and the accompanying keratin unfolding mechanics manuscript as appropriate for downstream work.

Validation Notes

The upload workflow checks unique record identifiers, expected Type I/Type II coverage, expected velocity groups for simulation outputs, sequence lengths, force-vector lengths, and internal PDB residue/SEQRES consistency. Per-record boolean fields retain known source-sequence discrepancy flags instead of silently overwriting them.

Citation

Please cite as:

@article{LuLeonforteBuehler2026,
    title={Comparative Molecular Dynamics Characterization of Hair Keratin Unfolding Mechanics},
    author={Wei Lu, Fabien Leonforte, Markus J. Buehler},
    journal={xxx},
    year={2026},
    url={http://XYZ.XYZ},
}