|
| |
Our general interests are mainly in the development of multi-scale
moldels and advanced sampling methods, driven by the study of
biomolecular structure, dynamics and function in biology and
medicine. A particular emphasis is on understanding how
intrinsic disorder of proteins mediates function and how such
functional mechanisms may be altered in human diseases. Another
focus is integrating simulation and experiemnt to understand the
mechanism and control of large-scale conformational transitions in
functional proteins and protein complexes including transmenbrane ion
channels and transporters. We have recently became very interested in
studying the driving forces, properties and mechanisms of protein and
nucleic acid condensates. Our research has been supported by NSF, NIH,
AHA, and various local and state funding agencies.
1. Multi-scale methods for simulating dynamic proteins, RNAs
and DNAs
A primary focus of our research is to develop efficient physics-based
models of biomolecules for describing their conformational
equilibria at atomistic and coarse-grained (CG) scales. All models
have been distributed in the widely available CHARMM/OpenMM software
packages.
Atomistic imiplicit solvent protein force fields:
the mean influence of solvent can be captured through direct
estimation of the solvation free energy, reducing the system size
~10-fold. Implicit solvent is considered a highly desirable
alternative to traditional explicit solvent. We
were among the first to demonstrate that many limitations of the
generalized Born (GB) implicit solvent could be addressed by
careful parameterization of physical parameters (e.g., atomic input
radii) together with the underlying protein force field (JACS 2006;
JCC 2017). We exploited enhanced sampling techniques and GPU
acceleration (Biomolecules 2021; see below) to
simulate conformational equilibria of carefully chosen model
peptides to achieve sufficient cancellation of errors, a strategy
has now become a standard practice in optimization of implicit as
well as explicit solvent protein force fields.
|
Intermediate resolution coarse-grained models: Efficient
coarse-grained (CG) models are often necessary to reach the length
and time scales of large-scale transitions, dynamic interactions and
phase transitions. The resolution in CG representation is critical for
adequate description of major driving forces underlying the system
of interest. We have developed a hybrid resolution (HyRes)
protein model for reliable description of local and long-range
structures of disordered proteins (PCCP 2017; JCIM 2022). The
backbone is represented at all-atom level, to capture
backbone interactions and transient structures, and side chains are
represented at intermediate resolution, allowing
semi-quantitative description of transient long-range orders. HyRes
has demonstrated successes in modeling dynamic protein interactions
as well as phase separation (see below). We are also developing
an intermediate resolution model for condensates of RNAs
(iConRNA) that can capture key local and long-range structure
features of dynamic RNAs and simulate their spontaneous phase
transitions with Mg2+ (PNAS 2025). iConRNA successfully
recapitulates experimentally observed lower critical solution
temperature (LCST) phase separation of model RNAs, and critically,
the nontrivial dependence on sequence, length, concentration, and
Mg2+ level.
|
Multi-scale enhanced sampling (MSES): Sufficient sampling
of relevant biomolecular conformations is crucial for achieving
quantitative and predictive simulations. We recently developed an
effective approach where efficient CG models are coupled with
atomistic ones to dramatically accelerate sampling at atomistic
level. The bias from the coupling is removed by Hamiltonian replica
exchange, thus allowing one to benefit simultaneously from faster
transitions of CG modeling and accuracy of atomistic force fields.
Initial application of MSES to nontrivial beta-hairpins shows that
MSES is up to >100-fold more efficient than temperature replica
exchange (T-RE). MSES is highly tunable, scalable to larger
proteins, and relatively robust against artifacts of the CG model
without the later dictating the conformational
diffusion. Furthermore, MSES does not require pre-specified reaction
coordinates or modifications to the atomistic
energy function. Further development of MSES is currently a key area
of focus for research in our lab.
|
(Spontaneous phase separation of GY23
Zhang et al, JACS 2024)
2. Biomolecular condensates: driving forces, mechanism and properties
Biomolecular condensates form and dissolve through spontaneous phase
separation, driven primarily by dynamic, multivalent biomolecules
such as RNAs and intrinsically disordered proteins (IDPs). These
condensates play crucial roles in a myriad of cellular processes,
including RNA storage and processing, stress response, metabolism,
and cellular signaling. Efficient coarse-grained (CG) models are
required to reach the length and time scales of
biomolecular phase transitions; they have been instrumental in
working hand-in-hand with theory and experiment to elucidate the
mechanism and regulation of the phase transition. Moving beyond the
popular one-bead per residue (or nucleotide) models, we are develop
intermediate resolution CG models such as HyRes and iConRNA (and
iConDNA) (JACS 2024; PNAS 2025) to achieve a superior balance
between computational efficiency and ability to capture the
interplay of diverse interactions and transient structures in
simulating protein and RNA dynamics and phase separation. Integrated
with atomistic simulations and experimental studies, we aim to
understand the driving forces and control of biomolecular
condensates as well as the properties and interaction of the
condensate with various small and macro-molecules in biology and
biomedical engineering.
3. Intrinsically disordered proteins: structure, function and disease
(Electrostatically accelerated coupled bindign and folding
of p27, Ganguly et al., JMB 2012)
Intrinsically disordered proteins (IDPs) are an important newly
recognized class of functional proteins that rely on a lack of stable
structures for function. They are highly abundant, play fundamental
roles in biology, and are extensively involved in human
diseases. We have been able to leverage our expertise
in advanced sampling, implicit solvent and NMR, and make progresses
toward establishing an effective computational framework for
reliable atomistic simulations of IDPs, and applying it to
understand the conformational properties of several IDPs and
determine how these properties may mediate IDP interaction and
regulation. A key lesson from our studies is that, although
preformed structures frequently exist in unbound IDPs, regulatory
IDPs tend to follow induced-folding like mechanisms. We
have shown that the somewhat surprising prevalence of
induced folding in IDPs is likely due to the constraint of fast
association required for cellular signaling and
regulation. Combining coarse-grained simulations and experiments, we
have further demonstrated that facile recognition of IDPs often involves a
novel electrostatically accelerated encounter and folding mechanism,
where long-range electrostatic forces not only accelerate protein
encounter, but also promote folding-competent encounter topologies
to enhance the efficiency of IDP folding upon encounter.
Besides efforts to understand the physical principles of IDP structure
and interaction, we have also been focusing on studies of the
structure-function-disease relationship of several important IDPs
including Bcl2 family proteins and tumor suppressor p53. In
particular, p53 is the most frequently mutated protein in
human cancers. Clinical studies have shown that the type of p53
mutation can be linked to cancer prognosis, response to treatment,
and patient survival. It is thus crucial to understand how various
mutations affect p53 at the molecular level, so as to understand the
biological outcomes and assess potential cancer intervention
strategies. The intrinsically disordered N-terminal transactivation
domain (TAD), in particular, is essential for p53's interactions
with multiple signaling networks that mediate the cellular response
to genotoxic stresses. We are combinging atomistic simulations, NMR
and other biophyscial approaches to determine the structural and
functional consequences of numerous cancer-associated mutations in
p53-TAD.
4. Gating and pore-sensor coupling mechanisms of channel proteins.
Ion channels facilitate the flow of ions through cell membranes via
opening of their pores in response to various electrical and
chemical signals. The transmembrane pore in ion channels generally
contain structures that control the types of permeating ions, known
as the selectivity filter, and those that open and close to control
the flow of ions, known as the gate. Upon sensing physiological
stimulations, the channel proteins change conformation to open the
gate and allow ion passage. In collaboration with various structural
biology and electrophysiology labs, we combine atomistic
simulations and biophysical experiments to understand the gating and
regulation of various ion channels and transporters including
calcium ativated potassium (BK) channels, TMEM16 ion channel and
lipid scramblase, and TRP family ion channels. An important recent
discovery is how hydrophobic dewetting of the inner pore provides a
novel gating mechansim in BK and TRPV4 channels.
5. Physics-augmented machine learning for
protein function and dynamics
(Reliable prediction of BK gating voltage)
Advances in GPU computing, sampling algorithms and empirical force
fields have continued to improve the ability of molecular simulations
to generate vast amount of quantitative data on biomolecular structural,
dynamic and energetic properties at atomistic details and up to
micorsecond timescales. Yet, these simulations remain far too short
to reach functional timescales, and reliable extraction of functional
mechanism remains a grant challenge in computational
biophysics. There is an urgent need and exciting opportunity to
leverage advances in data sciences to address this challenge. In
particular, large amounts of functional data have also accumulated for
many important proteins that probe how function may be perturbed by
mutation, binding, temperature etc. The molecular properties derived
from modeling and simulation should provide the physical basis of the observed
functional outputs. The bottleneck here is that the correlation is
complex, nonlinear and highly nontrivial to recover. We are interested
in learning how to integrate informational and
statistical approaches with molecular modeling and simulation to
reliably reconstruct the complex correlation between fundamental
molecular properties and functional outputs, as well as to inform what
and how new calculations and experiments should be performed such as
to iteratively improve the quality of the reconstruction (e.g., PLoS
Comput Biol 2023). We are also interested in designing and training
generative models for disordered protein ensembles based on
simulation, sequence evolution and structure database.
6. Computational modeling and design of biomimetic materials
Multi-scale modeling is a powerful approach to assist the
design and characterization of novel biomimetic materials, including
a new class of self-assembling branched amphiphilic peptide capsules
(BAPCs) (Langmuir 2016). These vesicles have unusual but highly
desirable properties that could address the shortcoming of
lipid-based carriers. These short vesicle-forming peptides (15-23
residues) can be synthesized economically and assembled easily by
users. In collaboration with the Thai lab at UMass, we are designing
synthetic polymers for protein binding and sensing. These polymers
are inspired by intrinsically disordered proteins that nature has
evolved for versatile protein recognition and regulation. We are
particularly interested in developing machine-learning-based design
methods leveraging chemical and physical data and molecular modeling.
Want to find out more about our research? Check out our latest
publications, or, simply drop by if you can! We are in LSL S(outh)585.
Back to Top
|