Structural
Plasticity Drives CDK1/CDK2 Isoform Selectivity: Atomistic Insights from
Molecular Dynamics
Ganga
Graziela Chrisna1, Chen Qu 2
1chrisnalove23@gmail.com, Zhejiang
University of Science and Technology, Hangzhou 310000, Zhejiang, China.
2chengqu@vip.126.com, Zhejiang
University of Science and Technology, Hangzhou 310000, Zhejiang, China.
Abstract
CDK1 and CDK2 are structurally
near-identical kinases whose ATP-binding pockets share a global backbone RMSD
of approximately 0.72 Å, yet certain inhibitors display up to 170-fold
selectivity for CDK2 over CDK1. Crystal structures alone cannot explain this
divergence. Here, we report 100 ns all-atom molecular dynamics simulations of
five CDK–inhibitor complexes involving three clinical-grade inhibitors,
Dinaciclib, AZD5438, and CGP74514A, benchmarked against experimental isothermal
titration calorimetry (ITC) data. Trajectory analysis reveals that isoform
selectivity is encoded in local pocket dynamics rather than global structural
differences. CDK2 undergoes productive induced-fit rearrangements, establishing
dense, persistent hydrogen-bonding networks and a tighter hydrophobic enclosure
of the ligand. CDK1 is conformationally rigid, prevents ordered water-mediated
bridging, and permits repeated solvent exposure of the bound inhibitor.
Per-residue energy decomposition and contact map analysis identify specific
hinge-region and PSTAIRE-helix residues as the primary determinants of
selectivity.
Keywords: CDK1; CDK2;
molecular dynamics; isoform selectivity; induced-fit; hydrogen bonding;
MM/PBSA; kinase inhibitors
1. Introduction
Cyclin-dependent kinases 1 and 2
(CDK1 and CDK2) occupy a central position in the regulation of the eukaryotic
cell cycle. CDK1, in complex with Cyclin A or Cyclin B, functions as the master
mitotic kinase responsible for nuclear envelope breakdown, spindle assembly,
and chromosome segregation [1]. CDK2, partnered with Cyclin
E or Cyclin A, governs the G1/S transition and sustains DNA replication. Both
kinases are validated oncology targets: aberrant CDK activity drives
uncontrolled proliferation across a wide spectrum of human cancers, and the
clinical success of CDK4/6 inhibitors, palbociclib, ribociclib, and
abemaciclib, has firmly established CDK inhibition as a viable precision
oncology strategy [2] [3].
The central challenge in
targeting CDK1 and CDK2 is selectivity. Their catalytic domains share
exceptionally high structural homology, with a global backbone RMSD of
approximately 0.72 Å, and the ATP-binding pockets that harbor virtually all
clinical inhibitors are nearly superimposable [4]. This conservation is not
coincidental; all kinases must recognize and utilize ATP as a phosphate donor,
a constraint that has been preserved over hundreds of millions of years of
evolution. The practical consequence for drug discovery is severe: most
inhibitors cannot distinguish between the two isoforms, yet targeting CDK1 in
normal tissues disrupts essential mitotic progression and causes dose-limiting
toxicities that derailed early pan-CDK inhibitor programs [5].
Paradoxically, experimental ITC
data demonstrate that certain inhibitors achieve striking selectivity despite
the structural near-identity of the two active sites. AZD5438, for instance,
binds cyclin-free CDK2 with a Kd of 26 ± 3 nM but achieves only 4,400 ± 3,200
nM affinity for cyclin-free CDK1, a 170-fold difference corresponding to a ΔΔG
of 2.87 kcal/mol. Dinaciclib exhibits a similar ~23-fold preference for CDK2.
Static crystal structures cannot account for this divergence; the active-site
contacts visible in co-crystal structures are largely identical across the two
isoforms. The missing variable is conformational dynamics [1].
Molecular dynamics (MD)
simulation provides an appropriate tool for accessing this dynamic dimension.
By propagating Newton's equations of motion over nanosecond-to-microsecond
timescales, MD reveals the conformational ensembles sampled by protein–ligand
complexes under physiological conditions, resolving the breathing motions, induced-fit
rearrangements, and water-network reorganizations that static structures
obscure [2] . Previous computational
studies have applied MM/PBSA end-point methods to CDK–inhibitor complexes and
demonstrated qualitative agreement with experimental affinity trends. Still, a
systematic, atomistic dissection of the structural and dynamic features
underpinning isoform selectivity across multiple inhibitors has not been
reported.
Here we present 100 ns all-atom
MD simulations of five CDK–inhibitor complexes: CDK1–Dinaciclib,
CDK2–Dinaciclib, CDK1–AZD5438, CDK2–AZD5438, and CDK2–CGP74514A, and report a
comprehensive trajectory analysis encompassing backbone dynamics, residue-level
flexibility, hydrogen-bond persistence, hydrophobic burial, and per-residue
interaction energetics. Our central finding is that CDK2 selectivity emerges
from the convergence of four linked properties: productive induced-fit
accommodation, persistent polar interaction networks, ordered water-mediated
bridging, and tighter hydrophobic enclosure. CDK1 fails on all four counts.
2. Materials and Methods
2.1 System Preparation
Initial atomic coordinates for
all five protein–ligand complexes were retrieved from the RCSB Protein Data
Bank: CDK1–Dinaciclib (PDB: 6GU6), CDK2–Dinaciclib (PDB: 4KD1), CDK1–AZD5438
(PDB: 6GU7), CDK2–AZD5438 (PDB: 6GUH), and CDK2–CGP74514A (PDB: 6GUK).
Crystallographic water molecules within 5 Å of the binding site were retained
to preserve water-mediated protein–ligand interactions. Each complex was placed
in a cubic TIP3P water box with a minimum buffer of 12 Å between any solute
atom and the box edge. Sodium and chloride ions were added to neutralize the
system charge and achieve a physiological ionic strength of 0.15 M [6] .
2.2 Force Field Parameterization
Protein components were parameterized
with the CHARMM36 all-atom additive force field. Small-molecule inhibitors were
parameterized using the OPLS-AA force field via the LigParGen web server. The
modified TIP3P water model was employed. All MD simulations were performed
using GROMACS 4.6.5 [7].
2.3 Simulation Protocol
Energy minimization was
conducted in two stages: 2,500 steps of steepest descent followed by 2,500
steps of conjugate gradient, first with positional restraints of 10.0
kcal/mol/Ų then without restraints. Systems were heated from 0 to 300 K over
50 ps in the NVT ensemble, followed by 500 ps NPT equilibration using the
Langevin thermostat and Monte Carlo barostat. Production runs of 100 ns each
were performed in the NPT ensemble. The SHAKE algorithm constrained all
hydrogen-bonded bonds with a 2 fs time step. Long-range electrostatics were
handled using the particle-mesh Ewald method with a 10 Å real-space cutoff [8].
2.4 Trajectory Analysis
Root-mean-square deviation
(RMSD) of protein Cα backbone atoms and ligand heavy atoms was computed
relative to the initial crystal structure with the binding site as reference
frame. Root-mean-square fluctuation (RMSF) of Cα atoms quantifies per-residue
flexibility. Hydrogen bonds were identified using a donor-acceptor distance
cutoff of 3.0 Å [9] . The solvent-accessible
surface area (SASA) of the inhibitor's heavy atoms was calculated using the
Shrake-Rupley method with a 1.4 Å water probe radius. Residue-level contact
maps were computed from the equilibrated 90–100 ns trajectory with direct
protein–ligand contacts defined at 3.5 Å. MM/PBSA binding free energies were
estimated using the gmmpbsa tool with entropic correction via the C2
approximation.
3. Results
3.1 Structural Contact Analysis and
Residue-Level Interaction Maps
Binding site contact maps
computed from the equilibrated 90–100 ns trajectory window reveal stark
differences in the persistence and extent of protein–ligand interactions
between CDK1 and CDK2 complexes (Figure 1). For the CDK2–AZD5438 complex, the
tightest binder experimentally (Kd = 26 ± 3 nM, Table 1), contact frequencies
exceed 0.8 for core ATP-binding pocket residues, including conserved
hinge-region and PSTAIRE helix residues [1]. In the CDK1–AZD5438 complex,
contact frequencies fall predominantly below 0.4, consistent with its 170-fold
weaker experimental affinity. The CDK2–Dinaciclib complex maintains persistent
contacts with hinge-region residues that are transient in the CDK1–Dinaciclib
complex, in agreement with the 23-fold experimental selectivity.
Crystallographic water molecules retained within 5 Å of the binding site
mediate additional indirect hydrogen bonds in CDK2 complexes; no equivalent
water-mediated contacts are observed in CDK1–AZD5438 [1].
Figure 1. Binding
site views and residue-level contact maps of CDK–inhibitor complexes.
Table 1. Experimental ITC binding data
for CDK–inhibitor complexes [1].
|
No. |
PDB |
Complex |
Kd (nM) |
ΔG (kcal/mol) |
|
1 |
6GU6 |
CDK1–Dinaciclib |
955 ± 246 |
−8.34 ± 0.20 |
|
2 |
4KD1 |
CDK2–Dinaciclib |
41 ± 14 |
−10.34 ± 0.37 |
|
3 |
6GU7 |
CDK1–AZD5438 |
4,400 ± 3,200 |
−7.76 ± 0.62 |
|
4 |
6GUH |
CDK2–AZD5438 |
26 ± 3 |
−10.63 ± 0.17 |
|
5 |
6GUK |
CDK2–CGP74514A |
715 ± 15 |
−8.50 ± 0.03 |
3.2 Global Structural Stability
Radius of gyration (Rg) values
remained stable throughout all 100 ns simulations, ranging from 1.92 to 2.06 Å
across all five complexes with no statistically significant differences between
CDK1 and CDK2 systems (p > 0.05, one-way ANOVA). Both kinases maintain their
native globular fold in the presence of all three inhibitors. This result
establishes that dramatic differences in inhibitor binding affinity are not the
consequence of global structural destabilization but arise exclusively from
local conformational dynamics within and around the ATP-binding pocket [2].
3.3 Ligand Accommodation and Induced-Fit
Dynamics
RMSD analysis of ligand and
protein backbone atoms over the 100 ns production trajectory reveals
fundamentally different accommodation mechanisms in CDK1 and CDK2 (Figure 2).
All systems achieve equilibrated backbone RMSD values between 0.15 and 0.35 nm
within the first 40–50 ns, validating the selection of the 90–100 ns window for
quantitative analysis [10].
CDK1–Dinaciclib maintains an
exceptionally stable ligand RMSD of ~0.1 nm throughout the trajectory,
indicating Dinaciclib is held rigidly in the CDK1 pocket without meaningful
conformational adaptation. By contrast, CDK2–Dinaciclib undergoes a notable
conformational transition at approximately 20 ns, with the ligand RMSD stabilizing
at ~0.2 nm a productive induced-fit rearrangement in which the CDK2 binding
pocket dynamically reorganizes to maximize intermolecular contacts.
CDK1–AZD5438 exhibits a transient ligand RMSD spike reaching 0.35 nm during the
initial 20 ns before settling into a stable but suboptimal binding mode.
CDK2–AZD5438 shows tightly synchronized ligand-protein backbone fluctuations
post-equilibration, indicative of a highly cooperative, stable binding
interface [10] [11].
Figure 2. RMSD time
evolution for all five CDK–inhibitor complexes. Protein Cα backbone RMSD (red)
and ligand heavy-atom RMSD (black) plotted relative to the initial crystal
structure over 100 ns. Horizontal dashed lines denote equilibrium thresholds
(2.0 Å protein; 1.0 Å ligand).
3.4 Regional Flexibility and Allosteric
Signatures
RMSF profiles of Cα atoms reveal
inhibitor-specific and isoform-specific flexibility patterns extending beyond
the immediate binding site, providing evidence for allosteric communication
between the inhibitor and distal protein regions (Figure 3). Overall, the
majority of residues in all complexes maintain RMSF values below 0.2 nm,
confirming core fold stability. Within CDK1 complexes, the Dinaciclib-bound
complex exhibits sharp fluctuations in residues 40–50, reaching ~0.55 nm, an
effect absent in the AZD5438-bound complex. CDK2 complexes display
inhibitor-specific allosteric signatures: residues 150–160 show elevated RMSF
in AZD5438-bound CDK2, while CGP74514A-bound CDK2 exhibits uniquely elevated
fluctuations around residues 70–80 [12] [13].
Figure 3. RMSF
profiles of Cα atoms for CDK1 and CDK2 complexes. (a) CDK1 bound to Dinaciclib
(black) and AZD5438 (red). (b) CDK2 bound to Dinaciclib (black), AZD5438 (red),
and CGP74514A (blue).
3.5 Hydrogen Bond Persistence as the
Selectivity Switch
Time-dependent analysis of
protein–ligand hydrogen bond counts over the 100 ns simulation reveals the
clearest correlation with experimental binding affinity among all trajectory
metrics examined (Figure 4). The CDK2–Dinaciclib complex maintains 3 to 5
persistent hydrogen bonds throughout the trajectory. CDK2–AZD5438 sustains 2 to
4 stable bonds post-equilibration. CDK2–CGP74514A oscillates between 1 and 3
bonds, consistent with its intermediate binding affinity [14] .
The contrast with CDK1 complexes
is stark. CDK1–AZD5438 (Kd = 4,400 ± 3,200 nM)
shows the most precarious binding, with the hydrogen bond count
repeatedly dropping to zero or one throughout the trajectory. Critically,
ordered crystallographic water molecules mediate stable, indirect hydrogen
bonds in CDK2 complexes, supplementing direct polar interactions. Greater
pocket flexibility in CDK1 displaces these bridging waters, eliminating this
supplementary stabilizing layer.
Figure 4. Time-dependent
hydrogen bond count between CDK isoforms and inhibitors over 100 ns.
CDK1–Dinaciclib (black), CDK2–Dinaciclib (red), CDK1–AZD5438 (blue),
CDK2–AZD5438 (dark cyan), CDK2–CGP74514A (magenta).
3.6 Hydrophobic Enclosure: SASA Analysis
SASA analysis of inhibitor heavy
atoms quantifies the degree of hydrophobic burial within the kinase-binding
pocket (Figure 5). For Dinaciclib, CDK2-bound complexes maintain a consistently
lower SASA, averaging ~0.8 nm², compared to CDK1-bound complexes, averaging
>1.0 nm². For AZD5438, the CDK1 complex shows multiple transient spikes
reaching 2.0 nm², while CDK2–AZD5438 achieves a stabilized SASA profile in the
final 40 ns of the trajectory. CDK2–CGP74514A displays marked SASA oscillations
with a notable decrease around 45 ns, followed by recovery, indicating dynamic
reorientation within the catalytic site [15].
Figure 5. Time-dependent
SASA of inhibitors over 100 ns. (a) Dinaciclib in CDK1 (black) and CDK2 (red).
(b) AZD5438 in CDK1 (black) and CDK2 (red). (c) CGP74514A in CDK2.
3.7 Binding Free Energy Decomposition
MM/PBSA binding free energy
calculations on snapshots from the equilibrated 90–100 ns window, with entropic
correction via the C2 approximation, provide a quantitative energetic context
for the structural observations above (Tables 2 and 3). The molecular mechanics
energy (ΔEmm) constitutes the predominant driving force across all complexes.
All complexes incur a substantial polar solvation penalty, while non-polar
solvation provides a modest stabilizing contribution, a profile characteristic
of ATP-competitive kinase inhibitor binding
[16] [17]. The CDK2–AZD5438 complex exhibits
the most favorable calculated binding free energy (−33.14 kcal/mol), consistent
with its highest experimental affinity.
The CDK2–Dinaciclib complex
incurs a substantial entropic penalty (−TΔS = 28.04 kcal/mol), reflecting the
significant loss of ligand conformational freedom upon the induced-fit pocket
rearrangement. While MM/PBSA, a well-documented limitation, overestimates
absolute binding free energies, it correctly identifies van der Waals and
electrostatic interactions as the core drivers of isoform selectivity.
Table 2. MM/PBSA energy decomposition
for CDK–inhibitor complexes.
|
Complex |
ΔEmm (kcal/mol) |
ΔGsol-polar (kcal/mol) |
ΔGsol-np (kcal/mol) |
|
CDK1–Dinaciclib |
−66.72 ± 4.01 |
30.67 ± 2.82 |
−5.96 ± 0.37 |
|
CDK2–Dinaciclib |
−69.48 ± 5.78 |
24.57 ± 3.21 |
−6.46 ± 0.41 |
|
CDK1–AZD5438 |
−56.41 ± 3.93 |
22.03 ± 1.76 |
−5.32 ± 0.40 |
|
CDK2–AZD5438 |
−56.49 ± 3.42 |
23.35 ± 1.88 |
−5.17 ± 0.34 |
|
CDK2–CGP74514A |
−53.34 ± 2.83 |
34.91 ± 5.38 |
−2.43 ± 0.24 |
Table 3. Predicted MM/PBSA binding free
energies versus experimental ITC values.
|
Complex |
ΔEbind |
−TΔSbind |
ΔGbind calc. |
ΔGexp (kcal/mol) |
|
CDK1–Dinaciclib |
−42.02 ± 3.21 |
13.50 ± 0.47 |
−28.52 ± 3.24 |
−8.15 ± 0.12 |
|
CDK2–Dinaciclib |
−44.91 ± 3.92 |
28.04 ± 0.98 |
−16.87 ± 4.04 |
−9.35 ± 0.16 |
|
CDK1–AZD5438 |
−34.38 ± 3.24 |
12.97 ± 0.45 |
−21.41 ± 3.27 |
−6.79 ± 0.42 |
|
CDK2–AZD5438 |
−33.14 ± 2.78 |
9.80 ± 0.37 |
−33.14 ± 2.78 |
−9.54 ± 0.07 |
|
CDK2–CGP74514A |
−20.86 ± 6.08 |
6.74 ± 0.20 |
−14.12 ± 6.08 |
−7.24 ± 0.01 |
4. Discussion
Our trajectory analysis
identifies four convergent mechanisms that together account for CDK2's superior
binding of all three inhibitors tested. First, CDK2 undergoes productive
induced-fit rearrangements: the ligand RMSD transitions observed at ~20 ns in
CDK2–Dinaciclib reflect dynamic pocket optimization rather than instability.
CDK1 lacks this plasticity; its rigid binding pocket forces the inhibitor into
a suboptimal pose. Second, CDK2 sustains dense, persistent hydrogen-bond
networks (3–5 bonds for Dinaciclib; 2–4 for AZD5438) that remain stable
throughout the trajectory. CDK1 networks are sparse and transient. Third,
ordered water molecules mediate stable, indirect hydrogen bonds in CDK2
complexes. Greater pocket flexibility in CDK1 displaces these bridging waters.
Fourth, CDK2 provides a tighter hydrophobic enclosure for all three inhibitors
(SASA ~0.2 nm² lower on average) [18] .
These four mechanisms operate
synergistically. The induced-fit rearrangement of CDK2's hinge region enables
the formation of a stable three-point hydrogen-bonding triad with Dinaciclib;
this triad stabilizes the ordered water network, and the overall pocket
reorganization creates a deeper hydrophobic cavity that buries the inhibitor
more completely. CDK1's rigidity prevents the first step in this cascade, and
all subsequent mechanisms fail as a result. This cascade model explains why the
experimental Kd differences are so large despite the structural near-identity
of the two active sites.
The design implications are
specific. Inhibitors targeting cyclin-free CDK2 should be optimized to engage
the dynamic hinge-region rearrangement, favoring scaffolds that establish a
three-point polar interaction network. The residues identified as persistent
contact partners in CDK2 but not CDK1, particularly PSTAIRE helix and
hinge-region residues, represent priority pharmacophore anchors. Limitations
include the use of classical, non-polarisable force fields and 100 ns
trajectories, which may miss slower conformational transitions on
microsecond-to-millisecond timescales.
5. Conclusion
This study provides an
atomistic, multi-metric characterization of the conformational and energetic
basis of CDK1/CDK2 isoform selectivity. Through 100 ns all-atom MD simulations
of five CDK–inhibitor complexes, we demonstrate that selectivity is encoded in
local pocket dynamics rather than global structural differences. CDK2 achieves
superior inhibitor binding through a convergence of four mechanisms:
induced-fit accommodation, persistent hydrogen-bond networks stabilized by
ordered water molecules, and tighter hydrophobic burial. CDK1 is
conformationally rigid and fails on all four counts. Per-residue contact
analysis identifies hinge-region and PSTAIRE-helix residues as primary
determinants of selectivity. These findings provide a mechanistic blueprint for
the rational design of next-generation CDK2-selective inhibitors.
References.
[1] S. Lapenna and A. Giordano, “Cell
cycle kinases as therapeutic targets for cancer,” Nat Rev Drug Discov,
vol. 8, no. 7, pp. 547–566, Jul. 2009, doi: 10.1038/nrd2907.
[2] M. P. Goetz et al., “MONARCH 3: Abemaciclib As
Initial Therapy for Advanced Breast Cancer,” JCO, vol. 35, no. 32, pp.
3638–3646, Nov. 2017, doi: 10.1200/JCO.2017.75.6155.
[3] R. S. Finn et al., “Palbociclib and Letrozole in
Advanced Breast Cancer,” N Engl J Med, vol. 375, no. 20, pp. 1925–1936,
Nov. 2016, doi: 10.1056/NEJMoa1607303.
[4] G. Manning, D. B. Whyte, R. Martinez, T. Hunter, and S.
Sudarsanam, “The Protein Kinase Complement of the Human Genome,” Science,
vol. 298, no. 5600, pp. 1912–1934, Dec. 2002, doi: 10.1126/science. 1075762.
[5] S. Haque et al., “Microbial dysbiosis and
epigenetics modulation in cancer development – A chemopreventive approach,” Seminars
in Cancer Biology, vol. 86, pp. 666–681, Nov. 2022, doi:
10.1016/j.semcancer.2021.06.024.
[6] M. P. Martin, S. H. Olesen, G. I. Georg, and E. Schönbrunn,
“Cyclin-Dependent Kinase Inhibitor Dinaciclib Interacts with the Acetyl-Lysine
Recognition Site of Bromodomains,” ACS Chem. Biol., vol. 8, no. 11, pp.
2360–2365, Nov. 2013, doi: 10.1021/cb4003283.
[7] J. Huang and A. D. MacKerell, “CHARMM36 all-atom additive
protein force field: Validation based on comparison to NMR data,” J. Comput.
Chem., vol. 34, no. 25, pp. 2135–2145, Sep. 2013, doi: 10.1002/jcc.23354.
[8] H. C. Andersen, “Rattle: A ‘velocity’ version of the shake
algorithm for molecular dynamics calculations,” Journal of Computational
Physics, vol. 52, no. 1, pp. 24–34, Oct. 1983, doi:
10.1016/0021-9991(83)90014-1.
[9] A. Shrake and J. A. Rupley, “Environment and exposure to
solvent of protein atoms. Lysozyme and insulin,” Journal of Molecular
Biology, vol. 79, no. 2, pp. 351–371, Sep. 1973, doi:
10.1016/0022-2836(73)90011-9.
[10] J. A. McCammon, B. R. Gelin, and M. Karplus, “Dynamics of
folded proteins,” Nature, vol. 267, no. 5612, pp. 585–590, Jun. 1977,
doi: 10.1038/267585a0.
[11] M. S. Mutentu, B. G. M. Horacio, and Y. Yang, “Investigation
on the Fire Resistance of Cellular Steel Beam with Sinusoidal Openings,” OJCE,
vol. 13, no. 04, pp. 637–663, 2023, doi: 10.4236/ojce.2023.134043.
[12] R. Nussinov and C.-J. Tsai, “Allostery in Disease and in Drug
Discovery,” Cell, vol. 153, no. 2, pp. 293–305, Apr. 2013, doi:
10.1016/j.cell.2013.03.034.
[13] Y. Wang, M. L. Diabakanga Batatana, and M. H. Bikoumou
Gambat, “Public perceptions of government policies to COVID-19: A comparative
study in six African countries,” Heliyon, vol. 10, no. 3, p. e24888,
Feb. 2024, doi: 10.1016/j.heliyon.2024.e24888.
[14] S. Genheden and U. Ryde, “The MM/PBSA and MM/GBSA methods to
estimate ligand-binding affinities,” Expert Opinion on Drug Discovery,
vol. 10, no. 5, pp. 449–461, May 2015, doi: 10.1517/17460441.2015.1032936.
[15] M. H. Bikoumou Gambat, H. Li, and J. Zhang, “Breaking
barriers to integrated digital Delivery (IDD) adoption in the construction
Industry: Evidence from five Central African countries,” Ain Shams
Engineering Journal, vol. 17, no. 7, p. 104216, Jul. 2026, doi:
10.1016/j.asej.2026.104216.
[16] R. Abel, L. Wang, E. D. Harder, B. J. Berne, and R. A.
Friesner, “Advancing Drug Discovery through Enhanced Free Energy Calculations,”
Acc. Chem. Res., vol. 50, no. 7, pp. 1625–1632, Jul. 2017, doi:
10.1021/acs.accounts.7b00083.
[17] S. A. Maxime, B. G. M. Horacio, and M. S. Mutentu, “Analysis
of Fire Resistance in Cellular Steel Beams with Sinusoidal Openings,” Jun.
2024, doi: 10.5281/ZENODO.11473823.
[18] S. Genheden and U. Ryde, “The MM/PBSA and MM/GBSA methods to
estimate ligand-binding affinities,” Expert Opinion on Drug Discovery,
vol. 10, no. 5, pp. 449–461, May 2015, doi: 10.1517/17460441.2015.1032936.
.jpg)
