14 examples of how LLMs can transform materials science and chemistry: a reflection on a large language model hackathon.

Large-language models (LLMs) such as GPT-4 caught the interest of many scientists. Recent studies suggested that these models could be useful in chemistry and materials science. To explore these possibilities, we organized a hackathon. This article chronicles the projects built as part of this hackathon. Participants employed LLMs for various applications, including predicting properties of molecules and materials, designing novel interfaces for tools, extracting knowledge from unstructured data, and developing new educational applications. The diverse topics and the fact that working prototypes could be generated in less than two days highlight that LLMs will profoundly impact the future of our fields. The rich collection of ideas and projects also indicates that the applications of LLMs are not limited to materials science and chemistry but offer potential benefits to a wide range of scientific disciplines.

Single-Use Capture Purification of Adeno-Associated Viral Gene Transfer Vectors by Membrane-Based Steric Exclusion Chromatography.

We present membrane-based steric exclusion chromatography (SXC) as a universal capture step for purification of adeno-associated virus (AAV) gene transfer vectors independent of their serotype and surface characteristics. SXC is performed by mixing an unpurified cell culture supernatant containing AAV particles with polyethylene glycol (PEG) and feeding the mixture onto a chromatography filter unit. The purified AAV particles are recovered by flushing the unit with a solution lacking PEG. SXC is an inexpensive single-use method that permits to concentrate, purify, and re-buffer AAV particles with yields >95% and >80% impurity clearance. SXC could theoretically be employed at industrial scales with units of nearly 20 m2.

Steric exclusion chromatography for purification of cell culture-derived influenza A virus using regenerated cellulose membranes and polyethylene glycol.

Steric exclusion chromatography has been used for the purification of proteins and bacteriophages using monoliths. The operation is carried out by mixing a crude sample containing the target species with a predetermined concentration and molecular weight of polyethylene glycol (PEG) and loading it onto a non-reactive hydrophilic surface. Product capture occurs by the mutual steric exclusion of PEG between the product and the matrix. Selectivity is significantly influenced by target product size. Product elution is achieved by decreasing the PEG concentration. In this study, a 75cm2 cellulose membrane adsorber was used for the purification of a clarified and inactivated influenza A virus broth produced in a 5L bioreactor using suspension Madin Darby canine kidney cells. Product recovery was above 95% based on hemagglutination activity and single radial immunodiffusion assays. Maximum depletion of double stranded host cell DNA and total protein was 99.7% and 92.4%, respectively. Purified virus particles showed no aggregation with a monodisperse peak around 84nm. 250mL of the clarified inactivated virus broth was purified within 40min. The surface area productivity based on the recovery of the viral hemagglutinin antigen was 28-50mgm-2h-1 depending on the feed and loading conditions.

Influence of the production system on the surface properties of influenza A virus particles.

In this study, influenza A/Puerto Rico/8/34 H1N1 virus particles (VP) produced in adherent and suspension Madin Darby canine kidney cells were investigated with a broad analytical toolbox to obtain more information on the VP's surface properties potentially affecting their aggregation behavior. First, differences in aggregation behavior were revealed by VP size distributions obtained via differential centrifugal sedimentation and confirmed by dynamic light scattering. The VP produced in adherent cells showed increased levels of aggregation in a 20 mM NaCl 10 mM Tris-HCl pH 7.4 low-salt buffer. This included the formation of multimers (dimers up to pentamers), whereas VP produced in suspension cells displayed no tendency toward aggregate formation. To investigate the cause of these differences in aggregation behavior, the VP samples were compared based on their zeta potential, their surface hydrophobicity, their lipid composition, and the N-glycosylation of their major VP surface protein hemagglutinin. The zeta potential and the hydrophobicity of the VP produced in the adherent cells was significantly decreased compared to the VP produced in the suspension cells. The lipid composition of both VP systems was approximately identical. The hemagglutinin of the VP produced in adherent cells included more of the larger N-glycans, whereas the VP produced in suspension cells included more of the smaller N-glycans. These results indicate that differences in the glycosylation of viral surface proteins should be monitored to characterize VP hydrophobicity and aggregation behavior, and to avoid aggregate formation and product losses in virus purification processes for vaccines and gene therapy.

Specific ion effects on the particle size distributions of cell culture-derived influenza A virus particles within the Hofmeister series.

Virus particle (VP) aggregation can have serious implications on clinical safety and efficacy of virus-based therapeutics. Typically, VP are suspended in buffers to establish defined product properties. Salts used to achieve these properties show specific effects in chemical and biological systems in a reoccurring trend known as Hofmeister series (HS). Hofmeister series effects are ubiquitous and can affect colloidal particle systems. In this study, influences of different ions (anions: SO4 2-, HPO4 2-, Cl-, Br-, NO3 -, I-; cations: K+, Na+, Li+, Mg2+, Ca2+) on particle size distributions of cell culture-derived influenza VP were investigated. For the experimental setup, influenza virus A/Puerto Rico/8/34 (H1N1) VP produced in adherent and suspension Madin Darby canine kidney cells were used. Inactivated and concentrated virus harvests were dialyzed against buffers containing the ions of interest, followed by differential centrifugal sedimentation to measure particle size distributions. VP from both cell lines showed no aggregation over a wide range of buffers containing different salts in concentrations ≥60 mM. However, when dialyzed to low salt or Ca2+ buffers, VP produced in adherent cells showed increased aggregation compared to VP produced in suspension cells. Additionally, changes in VP diameters depending on specific ion concentrations were observed that partially reflected the HS trend.

A cell culture-derived whole virus influenza A vaccine based on magnetic sulfated cellulose particles confers protection in mice against lethal influenza A virus infection.

Downstream processing and formulation of viral vaccines employs a large number of different unit operations to achieve the desired product qualities. The complexity of individual process steps involved, the need for time consuming studies towards the optimization of virus yields, and very high requirements regarding potency and safety of vaccines results typically in long lead times for the establishment of new processes. To overcome such obstacles, to enable fast screening of potential vaccine candidates, and to explore options for production of low cost veterinary vaccines a new platform for whole virus particle purification and formulation based on magnetic particles has been established. Proof of concept was carried out with influenza A virus particles produced in suspension Madin Darby canine kidney (MDCK) cells. The clarified, inactivated, concentrated, and diafiltered virus particles were bound to magnetic sulfated cellulose particles (MSCP), and directly injected into mice for immunization including positive and negative controls. We show here, that in contrast to the mock-immunized group, vaccination of mice with antigen-loaded MSCP (aMSCP) resulted in high anti-influenza A antibody responses and full protection against a lethal challenge with replication competent influenza A virus. Antiviral protection correlated with a 400-fold reduced number of influenza nucleoprotein gene copies in the lungs of aMSCP immunized mice compared to mock-treated animals, indicating the efficient induction of antiviral immunity by this novel approach. Thus, our data proved the use of MSCP for purification and formulation of the influenza vaccine to be fast and efficient, and to confer protection of mice against influenza A virus infection. Furthermore, the method proposed has the potential for fast purification of virus particles directly from bioreactor harvests with a minimum number of process steps towards formulation of low-cost veterinary vaccines, and for screening studies requiring fast purification protocols.

Comparison of Influenza Virus Particle Purification Using Magnetic Sulfated Cellulose Particles with an Established Centrifugation Method for Analytics.

A method for the purification of influenza virus particles using novel magnetic sulfated cellulose particles is presented and compared to an established centrifugation method for analytics. Therefore, purified influenza A virus particles from adherent and suspension MDCK host cell lines were characterized on the protein level with mass spectrometry to compare the viral and residual host cell proteins. Both methods allowed one to identify all 10 influenza A virus proteins, including low-abundance proteins like the matrix protein 2 and nonstructural protein 1, with a similar impurity level of host cell proteins. Compared to the centrifugation method, use of the novel magnetic sulfated cellulose particles reduced the influenza A virus particle purification time from 3.5 h to 30 min before mass spectrometry analysis.