Search for: All records

Abstract Adenosine-to-inosine (A-to-I) messenger RNA (mRNA) editing can affect the sequence and function of translated proteins and has been extensively investigated in eukaryotes. However, the prevalence of A-to-I mRNA editing in bacteria, its governing regulatory principles, and its biological significance are poorly understood. Here, we show that A-to-I mRNA editing occurs in hundreds of transcripts across dozens of gammaproteobacterial species, with most edits predicted to recode protein sequences. Furthermore, we reveal conserved regulatory determinants controlling editing across gammaproteobacterial species. Using Acinetobacter baylyi as a model, we show that mutating TadA, the mediating enzyme, reduces editing across all sites. Conversely, overexpressing TadA resulted in the editing of >300 transcripts, attesting to the editing potential of TadA. Notably, we show for the first time, at the protein level, that normal levels of A-to-I mRNA editing lead to wild-type bacteria expressing two protein isoforms from a single gene. Finally, we show that a TadA mutant with deficient editing activity does not grow at high temperatures, suggesting that RNA editing has a functional role in bacteria. Our work reveals that A-to-I mRNA editing in bacteria is widespread and has the potential to reshape the bacterial transcriptome and proteome. 

Synopsis In the era of big data, ecological research is experiencing a transformative shift, yet big-data advancements in thermal ecology and the study of animal responses to climate conditions remain limited. This review discusses how big data analytics and artificial intelligence (AI) can significantly enhance our understanding of microclimates and animal behaviors under changing climatic conditions. We explore AI’s potential to refine microclimate models and analyze data from advanced sensors and camera technologies, which capture detailed, high-resolution information. This integration can allow researchers to dissect complex ecological and physiological processes with unprecedented precision. We describe how AI can enhance microclimate modeling through improved bias correction and downscaling techniques, providing more accurate estimates of the conditions that animals face under various climate scenarios. Additionally, we explore AI’s capabilities in tracking animal responses to these conditions, particularly through innovative classification models that utilize sensors such as accelerometers and acoustic loggers. For example, the widespread usage of camera traps can benefit from AI-driven image classification models to accurately identify thermoregulatory responses, such as shade usage and panting. AI is therefore instrumental in monitoring how animals interact with their environments, offering vital insights into their adaptive behaviors. Finally, we discuss how these advanced data-driven approaches can inform and enhance conservation strategies. In particular, detailed mapping of microhabitats essential for species survival under adverse conditions can guide the design of climate-resilient conservation and restoration programs that prioritize habitat features crucial for biodiversity resilience. In conclusion, the convergence of AI, big data, and ecological science heralds a new era of precision conservation, essential for addressing the global environmental challenges of the 21st century. 

A major risk of using language models in practical applications is their tendency to hallucinate incorrect statements. Hallucinations are often attributed to knowledge gaps in LMs, but we hypothesize that in some cases, when justifying previously generated hallucinations, LMs output false claims that they can separately recognize as incorrect. We construct three question-answering datasets where ChatGPT and GPT-4 often state an incorrect answer and offer an explanation with at least one incorrect claim. Crucially, we find that ChatGPT and GPT-4 can identify 67% and 87% of their own mistakes, respectively. We refer to this phenomenon as hallucination snowballing: an LM over-commits to early mistakes, leading to more mistakes that it otherwise would not make. 

Habitat loss poses a major threat to global biodiversity. Many studies have explored the potential damages of deforestation to animal populations but few have considered trees as thermoregulatory microhabitats or addressed how tree loss might impact the fate of species under climate change. Using a biophysical approach, we explore how tree loss might affect semi-arboreal diurnal ectotherms (lizards) under current and projected climates. We find that tree loss can reduce lizard population growth by curtailing activity time and length of the activity season. Although climate change can generally promote population growth for lizards, deforestation can reverse these positive effects for 66% of simulated populations and further accelerate population declines for another 18%. Our research underscores the mechanistic link between tree availability and population survival and growth, thus advocating for forest conservation and the integration of biophysical modelling and microhabitat diversity into conservation strategies, particularly in the face of climate change. 

In recent years, domains such as natural language processing and image recognition have popularized the paradigm of using large datasets to pretrain representations that can be effectively transferred to downstream tasks. In this work we evaluate how such a paradigm should be done in imitation learning, where both pretraining and finetuning data are trajectories collected by experts interacting with an unknown environment. Namely, we consider a setting where the pretraining corpus consists of multitask demonstrations and the task for each demonstration is set by an unobserved latent context variable. The goal is to use the pretraining corpus to learn a low dimensional representation of the high dimensional (e.g., visual) observation space which can be transferred to a novel context for finetuning on a limited dataset of demonstrations. Among a variety of possible pretraining objectives, we argue that inverse dynamics modeling – i.e., predicting an action given the observations appearing before and after it in the demonstration – is well-suited to this setting. We provide empirical evidence of this claim through evaluations on a variety of simulated visuomotor manipulation problems. While previous work has attempted various theoretical explanations regarding the benefit of inverse dynamics modeling, we find that these arguments are insufficient to explain the empirical advantages often observed in our settings, and so we derive a novel analysis using a simple but general environment model.