Search for: All records

Existing efforts on quantifying privacy implications for large language models (LLMs) solely focus on measuring leakage of training data. In this work, we shed light on the often-overlooked interactive settings where an LLM receives information from multiple sources and generates an output to be shared with other entities, creating the potential of exposing sensitive input data in inappropriate contexts. In these scenarios, humans nat- urally uphold privacy by choosing whether or not to disclose information depending on the context. We ask the question “Can LLMs demonstrate an equivalent discernment and reasoning capability when considering privacy in context?” We propose CONFAIDE, a benchmark grounded in the theory of contextual integrity and designed to identify critical weaknesses in the privacy reasoning capabilities of instruction-tuned LLMs. CONFAIDE consists of four tiers, gradually increasing in complexity, with the final tier evaluating contextual privacy reasoning and theory of mind capabilities. Our experiments show that even commercial models such as GPT-4 and ChatGPT reveal private information in contexts that humans would not, 39% and 57% of the time, respectively, highlighting the urgent need for a new direction of privacy-preserving approaches as we demonstrate a larger underlying problem stemmed in the models’ lack of reasoning capabilities. 

Tyrosinase, an important oxidase involved in the primary immune response in humans, can sometimes become problematic as it can catalyze undesirable oxidation reactions. Therefore, for decades there has been a strong pharmaceutical interest in the discovery of novel inhibitors of this enzyme. Recent studies have also indicated that tyrosinase inhibitors can potentially be used in the treatment of melanoma cancer. Over the years, many new tyrosinase inhibitors have been discovered from various natural sources; however, marine natural products (MNPs) have contributed only a small number of promising candidates. Therefore, in this study we focused on the discovery of new MNP tyrosinase inhibitors of marine cyanobacterial and algal origins. A colorimetric tyrosinase inhibitory assay was used to screen over 4,500 marine extracts against mushroom tyrosinase ( A. bisporus ). Our results revealed that scytonemin monomer (ScyM), a pure compound from our compound library and also the monomeric last-step precursor in the biosynthesis of the well-known cyanobacterial sunscreen pigment “scytonemin,” consistently showed the highest tyrosinase inhibitory score. Determination of the half maximal inhibitory concentration (IC 50 ) further indicated that ScyM is more potent than the commonly used commercial inhibitor standard “kojic acid” (KA; IC 50 of ScyM: 4.90 μM vs. IC 50 of KA: 11.31 μM). After a scaled-up chemical synthesis of ScyM as well as its O -methyl analog (ScyM-OMe), we conducted a series of follow-up studies on their structures, inhibitory properties, and mode of inhibition. Our results supported ScyM as the second case ever of a novel tyrosinase inhibitory compound based on a marine cyanobacterial natural product. The excellent in vitro performance of ScyM makes it a promising candidate for applications such as a skin-whitening agent or an adjuvant therapy for melanoma cancer treatment. 

Specialized or secondary metabolites are small molecules of biological origin, often showing potent biological activities with applications in agriculture, engineering and medicine. Usually, the biosynthesis of these natural products is governed by sets of co-regulated and physically clustered genes known as biosynthetic gene clusters (BGCs). To share information about BGCs in a standardized and machine-readable way, the Minimum Information about a Biosynthetic Gene cluster (MIBiG) data standard and repository was initiated in 2015. Since its conception, MIBiG has been regularly updated to expand data coverage and remain up to date with innovations in natural product research. Here, we describe MIBiG version 4.0, an extensive update to the data repository and the underlying data standard. In a massive community annotation effort, 267 contributors performed 8304 edits, creating 557 new entries and modifying 590 existing entries, resulting in a new total of 3059 curated entries in MIBiG. Particular attention was paid to ensuring high data quality, with automated data validation using a newly developed custom submission portal prototype, paired with a novel peer-reviewing model. MIBiG 4.0 also takes steps towards a rolling release model and a broader involvement of the scientific community. MIBiG 4.0 is accessible online at https://mibig.secondarymetabolites.org/.