Abstract:
Biomolecular phase separation has been recently recognized to be an important mechanism of subcellular compartmentalization, and it's closely associated with numerous cellular activities and human diseases. The phase separation is driven by multivalent interactions mediated frequently by intrinsically disordered proteins (IDPs) and dynamic RNAs. Intense attentions have been devoted to combine theory, simulation and experiment to understand sequence-specific phase separation and material properties of biomolecular condensates. However, existing coarse-grained protein and RNA models are severely limited in their ability to capture the diverse interactions and complex conformational properties of these biomolecules in phase separation. In this talk, I will introduce a hybrid resolution (HyRes) protein model for accurate description of the backbone and transient secondary structures of IDPs in phase separation. I will also discuss the development of a new RNA model that is capable of describing the dynamic structures of nontrivial RNAs and capturing sequence and length-dependence of their phase separation. We believe that these new models represents important advances in accurate and efficient molecular simulation of biomolecular phase separation, which can greatly enable mechanistic studies and rational biomaterial engineering. The ability to model complex and dynamic structure and interactions of IDPs and RNAs can also enrich the experimental data and help overcome data scarcity in future machine learning efforts towards predictive and generative modeling of biomolecular condensates.
Bio:
Jianhan Chen, Ph.D., is a Professor in the Department of Chemistry and Department Biochemistry and Molecular Biology at University of Massachusetts, Amherst. His research program focuses on the development of theoretical and computational methods and application of these methods to advance our understanding of biophysical, biochemical and biomedical problems. Key problems currently under study in Chen’s lab include intrinsically disordered proteins in biology and diseases, protein amyloid formation in neurogenerative diseases, self-assembling peptide vesicles for drug delivery, and transmembrane ion channel protein activation and regulation. Chen is a recipient of the NSF Career Award (2010), ACS Outstanding Junior Faculty Award (2011), the Outstanding Research Award in Research from UMass College of Natural Sciences (2022), and the NIH/NIGMS MIRA award (2022). He received his BS from the University of Science and Technology of China in 1998 and Ph.D. in Chemical and Material Physics from the University of California Irvine in 2002.
Summary
Liquid-Liquid Phase Separation (LLPS)
Thermodynamically-driven reversible de-mixing of (binary) solution into 2 different liquid phases
Many molecules tend to stick together with others of their own kind
Higher temperature -> molecules are forced to jump around the substance, away from their own kind, don’t form clumps
Higher density -> many travel paths are blocked, makes it more likely that molecules stay in place, form clumps (liquid droplets) of their own kind
Key interest: protein solutions
This process occurs in cells, create compartments/droplets within them
There are many biological processes that involve packaging of many protein molecules into droplets: Biological Condensates
Major drivers
Intrinsically Disordered Proteins (IDPs)
RNA
Intrinsically Disordered Proteins
Lack stable tertiary structure
⅓ of human genome, ⅔ of cancer-associated proteins
Described as dynamics and likely heterogeneous assemblies
Fold upon binding, self-assembly, phase separation
Physics-based Molecular Modeling and Simulation
Components:
Energy Function (force field)
Calculation of Dynamics (sampling)
Can be modeled at many levels of detail/abstraction
Classical Energy functions from underlying quantum dynamics
Dynamics:
Molecular Dynamics (MD): mechanistic motion of molecules
Monte Carlo sampling of likelihood of various structural moves based on transition energy
MD tends to be more efficient
Energy function -> Trajectory of conformation (probability distribution of possibilities) -> Free energy of each conformation -> Thermodynamic properties
But, extremely expensive
Timestep 1-2 fs (target timescale >> μs)
>>105 atoms
Molecular simulations of LLPS: Cα-model
Take a protein, focus on the shape of its amino acid backbone
Break it up into a sequence of segments that are treated as coarse “balls”
Interactions between these balls is modeled like in molecular dynamics, just coarser
Choice of this coarsening is the key challenge: major focus of research: manual/intuition, physics-based, data-driven
CALVADOS protein model
Bayesian Parameter-Learning Procedure
Chose different coarsenings, try them out in simulation, until they find the interaction parameters that consistently produces accurate approximations of coarse model vs fine-grained model
ALBATROSS: ML Prediction of IDP Dimensions
Train a Recurrent Neural Net on a Cα-model
Use the neural network for design
idpSAM: GEnerative Modeling of IDP Ensembles
Use atomic model to generate many ensembles
Train generative model
Generate realistic new ensembles from model -> sample conformational space
Major challenge:
Limited accuracy of trajectories from atomistic simulation because it is so expensive to run at full accuracy
Opportunity to use ML to speed up atomistic simulations
Protein Backbone Structures and Interactions
IDPS are not always random coils!
Transient structures are an important part of how consensates of IDPs form
Ongoing work: Hybrid Resolution Protein Model