לוח שנה משולב

Considerable efforts are devoted by living creatures to stabilization and preservation of dry proteins and other macromolecules. These efforts are echoed by attempts directed toward development of new, greener, and more effective preservation technologies, including attempts to extend food shelf life and to ehnace organ storage. I will describe our work to unravel the solvation and stabilization molecular mechanisms in two examples: imbedding proteins in a glassy matrix of sugar, and macromolecular solvation in deep eutectic solvents that are (almost) non-aqueous yet biologically compatible.

Zoom Link: https://weizmann.zoom.us/j/99290579488?pwd=cUIyV05SMUQ0VDErNUtma1RTL3BIQT09


In the last decade, halide perovskites (HaPs) have developed as promising new materials for a
wide range of optoelectronic applications, notably solar energy conversion. Although their
technology has advanced rapidly towards high solar energy conversion efficiency and
advantageous optoelectronic properties, many of their properties are still largely unknown from
a basic scientific standpoint. Due to the highly dynamical nature of HaPs, one of the main
avenues for basic science research is the use of molecular dynamics (MD) simulations, which
provide a full atomistic picture of those materials. One of the main limiting factors for such
analysis is the time scale of the MD simulation. Because of the complexity of the HaP system,
classical force field approaches do not yield satisfactory results and the most widely used force
calculation approach is based on first-principles, namely on density functional theory (DFT).
In recent years, a new type of force calculation approach has emerged, which is machine
learned force fields (MLFF). These methods are based on machine learning (ML) algorithms.
Their wide spread use is enabled by the ever-increasing computational power and by the
availability of large-scale shared repositories of scientific data. Here, we have applied one
MLFF algorithm, known as domain machine learning (GDML). After training a MLFF based
on the GDML model, we observed that the MLFF fails in a dynamical setting while still
showing low testing error. This has been found to be due to lack of full coverage of the
simulation phase space. To address this issue, we have suggested the hybrid temperature
ensemble (HTE) approach, where we create rare events that are training samples on the edge
of the phase space. We achieve this by combing MD trajectories from a range of temperatures
to a single dataset. The MLFF model, trained on the HTE dataset, showed increasing accuracy
during the training process, while being dynamically stable for a long duration of MD
simulation. The trained MLFF model also exhibited high accuracy for long-term simulations,
showing remaining errors of the same magnitude of inherent errors in DFT calculation.

Phase separation is generally a thermodynamic process in which a mixture reaches its lowest free energy state by self-assembling into meso- (or macro-) scale regions that are concentrated or dilute in a given molecular component. Familiar examples include the immiscibility of water and oil, the demixing of metal atoms in alloys, and the mesoscale formation of emulsions such as milk or paint. The fundamental physics behind both the equilibrium and non-equilibrium aspects of phase separation are well understood and this talk will begin with a brief review of those. A rapidly growing body of experiments suggests that phase separation is responsible for the formation of membraneless domains (also known as biomolecular condensates, with length scales on the order of microns) in biological cells. These compartments allow the cell to organize itself in space and can promote or inhibit biochemical reactions, provide regions in which macromolecular assemblies can form, or control the spatial organization of DNA (assembled with proteins as chromatin) in the cell nucleus. I will review some recent examples based on experiments done at the Weizmann Institute on phase separation of proteins and of chromatin in the nucleus and show how physics theory has led to their understanding. In the latter case, a new paradigm is emerging in which the genetic material is not necessarily uniformly distributed within the nucleus but separated into domains which in some cases, have a complex, “marshland”, mesoscale structure. But while many of the equilibrium aspects can be at least semi-quantitatively understood by extensions of statistical physics, biological systems often do not have constant overall compositions as is the case in the examples of oil-water, alloys and emulsions; for example, over time, the cell produces and degrades many proteins. The recent understanding of such strongly non-equilibrium effects has informed the theoretical physics of phase separation and has allowed us to establish a framework in which biological noise can be included.
* Collaborations: Omar Arana-Adame, Gaurav Bajpai, Dan Deviri, Amit Kumar (Dept. Chemical and Biological Physics), group of Emmanuel Levy (Dept. Structural Biology) and group of Talila Volk (Dept. Molecular Genetics)