Search for: All records

Complex coacervation is an associative liquid–liquid phase separation phenomenon that takes place due to the electrostatic complexation of oppositely-charged polyelectrolytes and the entropic gains associated with the release of bound counterions and rearrangement of solvent. The aqueous nature of coacervation has resulted in its broad use in systems requiring high biocompatibility. The significance of electrostatic interactions in coacervates has meant that studies investigating the phase behaviors of these systems have tended to focus on parameters such as the charge stoichiometry of the polyions, the solution pH, and the ionic strength. However, the equilibrium that exists between the polymer-rich coacervate phase and the polymer-poor supernatant phase represents a balance among attractive electrostatic interactions and excluded volume repulsions as well as osmotic pressure effects. As such, we hypothesize that it should be possible to tune coacervate phase behavior via the addition of non-electrostatic excipients which would partition between the two phases and potentially alter both the solvent quality and the osmotic pressure balance. In particular, our work focuses on small molecule excipients such as sugars, amino acids, and other additives that have a history of use in vaccine formulation. We quantified the ability of these excipients to partition into the coacervate phase, and their potential for destabilizing the phase separation. Furthermore, we demonstrate that these additives can be combined with complex coacervation in the context of a virus formulation. 

Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks (Qi et al., 2023)– a few harmful data mixed in the fine-tuning dataset can break the LLMs’s safety alignment. While several defenses have been proposed, our evaluation shows that existing defenses fail when some specific training hyper-parameters are chosen – a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains agnostic to the training hyper-parameters in the fine-tuning stage. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks. 

SARS-CoV-2, the cause of COVID-19, is a new, highly pathogenic coronavirus, which is the third coronavirus to emerge in the past 2 decades and the first to become a global pandemic. The virus has demonstrated itself to be extremely transmissible and deadly. Recent data suggest that a targeted approach is key to mitigating infectivity. Due to the proliferation of cataloged protein and nucleic acid sequences in databases, the function of the nucleic acid, and genetic encoded proteins, we make predictions by simply aligning sequences and exploring their homology. Thus, similar amino acid sequences in a protein usually confer similar biochemical function, even from distal or unrelated organisms. To understand viral transmission and adhesion, it is key to elucidate the structural, surface, and functional properties of each viral protein. This is typically first modeled in highly pathogenic species by exploring folding, hydrophobicity, and isoelectric point (IEP). Recent evidence from viral RNA sequence modeling and protein crystals have been inadequate, which prevent full understanding of the IEP and other viral properties of SARS-CoV-2. We have thus experimentally determined the IEP of SARS-CoV-2. Our findings suggest that for enveloped viruses, such as SARS-CoV-2, estimates of IEP by the amino acid sequence alone may be unreliable. We compared the experimental IEP of SARS-CoV-2 to variants of interest (VOIs) using their amino acid sequence, thus providing a qualitative comparison of the IEP of VOIs. 

In this study, we describe reducing the moisture vapor transmission through a commercial polymer bag material using a silicon-incorporated diamond-like carbon (Si-DLC) coating that was deposited using plasma-enhanced chemical vapor deposition. The structure of the Si-DLC coating was analyzed using scanning electron microscopy, Raman spectroscopy, X-ray photoelectron spectroscopy, energy-dispersive X-ray spectroscopy, selective area electron diffraction, and electron energy loss spectroscopy. Moisture vapor transmission rate (MVTR) testing was used to understand the moisture transmission barrier properties of Si-DLC-coated polymer bag material; the MVTR values decreased from 10.10 g/m2 24 h for the as-received polymer bag material to 6.31 g/m2 24 h for the Si-DLC-coated polymer bag material. Water stability tests were conducted to understand the resistance of the Si-DLC coatings toward moisture; the results confirmed the stability of Si-DLC coatings in contact with water up to 100 °C for 4 h. A peel-off adhesion test using scotch tape indicated that the good adhesion of the Si-DLC film to the substrate was preserved in contact with water up to 100 °C for 4 h.