NLM IRP Seminar Schedule

Virus discovery in the era of massive metagenomic sequencing

Accumulation of massive amount of non-targeted sequencing data allows to reverse traditional virus discovery pathway. Classically, viruses were discovered as disease agents, isolated, sequenced, and analyzed. Later, similarities between these sequences were built into virus classification and given an evolutionary perspective. Nowadays, it became possible to discover previously unknown and undetected viruses directly from (meta)genomic sequences. Annotation is heavily assisted by availability of a large amount of related virus sequences, which increase the sensitivity and reduces dependence on external libraries of known domains and functions. This also facilitates classification and evolutionary reconstruction concurrent with the discovery.
I illustrate this using the discovery of mriyaviruses (proposed class (“Mriyaviricetes”), a group of small relatives of giant viruses (Nucleocytoviricota). The most intriguing feature of “Mriyaviricetes” is their putative ancestral status with respect to previously described Nucleocytoviricota, as indicated by their deep placement in phylogenetic trees of the conserved proteins and by comparison of the major capsid protein structures. Analysis of proteins encoded in mriyavirus genomes suggests that they replicate their genome via the rolling circle mechanism that so far was not described for members of Nucleocytoviricota. Further expansion of the “Mriyaviricetes” through extended metagenome mining can be expected to further clarify and solidify the scenario for the origin and evolution of the phylum Nucleocytoviricota and viral gigantism.

TBD

Enhancing Heart Failure Prediction through LLM-backed Doctor Simulation

Accurate prediction systems are highly sought after in clinical care due to their potential benefits, such as aiding clinicians in decision-making, reducing costs, and enabling personalized treatment. However, obtaining annotated data for clinical risk prediction, particularly temporal information, is often expensive and time-consuming. In this paper, we focus on leveraging large language models (LLMs) to extract case reports and generate clinical events with associated timestamps from clinical texts, thereby reducing the costs of patient history collection and annotation. Specifically, we utilize GPT-4 to simulate a physician reading clinical texts and generating clinical event-timestamp pairs. These generated pairs are then used to perform a two-stage fine-tuning of LLaMA 2 on both generated and real data for heart failure prediction. Experimental results on a large EHR dataset, comprising 14 million visits by 263,000 patients over an 8-year period, demonstrate that our model, trained with generated clinical event-timestamp pairs, outperforms the baseline model trained solely on real data. Moreover, our approach is adaptable to other clinical risk prediction tasks in real-world settings.