Search NASA⌕ Search

SEARCH · Search NASA

Results for “annotations”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 469 records · Page 26

Learning from crowds with variational Gaussian processes

Solving a supervised learning problem requires to label a training set. This task is traditionally performed by an expert, who provides a label for each sample. The proliferation of social web services (e.g., Amazon Mechanical Turk) has introduced an alternative crowdsourcing approach. Anybody with a computer can register in one of these services and label, either partially or completely, a dataset. The effort of labeling is then shared between a great number of annotators. However, this approach introduces scientifically challenging problems such as combining the unknown expertise of the annotators, handling disagreements on the annotated samples, or detecting the existence of spammer and adversarial annotators. All these problems require probabilistic sound solutions which go beyond the naive use of majority voting plus classical classification methods. In this work we introduce a new crowdsourcing model and inference procedure which trains a Gaussian Process classifier using the noisy labels provided by the annotators. Variational Bayes inference is used to estimate all unknowns. The proposed model can predict the class of new samples and assess the expertise of the involved annotators. Moreover, the Bayesian treatment allows for a solid uncertainty quantification. Since when predicting the class of a new sample we might have access to some annotations for it, we also show how our method can naturally incorporate this additional information. Furthermore, a comprehensive experimental section evaluates the proposed method with synthetic and real experiments, showing that it consistently outperforms other state-of-the-art crowdsourcing approaches.

97 MATHEMATICS AND COMPUTING↗

Covering complete proteomes with X-ray structures: a current snapshot

The current and the attainable coverage by X-ray structures of proteins and their functions on the scale of the ‘protein universe’ are estimated. A detailed analysis of the coverage across nearly 2000 proteomes from all superkingdoms of life and functional annotations is performed, with particular focus on the human proteome and the family of GPCR proteins. Structural genomics programs have developed and applied structure-determination pipelines to a wide range of protein targets, facilitating the visualization of macromolecular interactions and the understanding of their molecular and biochemical functions. The fundamental question of whether three-dimensional structures of all proteins and all functional annotations can be determined using X-ray crystallography is investigated. A first-of-its-kind large-scale analysis of crystallization propensity for all proteins encoded in 1953 fully sequenced genomes was performed. It is shown that current X-ray crystallographic knowhow combined with homology modeling can provide structures for 25% of modeling families (protein clusters for which structural models can be obtained through homology modeling), with at least one structural model produced for each Gene Ontology functional annotation. The coverage varies between superkingdoms, with 19% for eukaryotes, 35% for bacteria and 49% for archaea, and with those of viruses following the coverage values of their hosts. It is shown that the crystallization propensities of proteomes from the taxonomic superkingdoms are distinct. The use of knowledge-based target selection is shown to substantially increase the ability to produce X-ray structures. It is demonstrated that the human proteome has one of the highest attainable coverage values among eukaryotes, and GPCR membrane proteins suitable for X-ray structure determination were determined.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Year 2 Final Report - Protein Function Prediction Platform (v.0.5)

Upon completion of our second year of development in a 3-year development cycle, we have completed a prototype protein structure-function annotation and function prediction system: Protein Function Prediction (PFP) platform (v.0.5). We have met our milestones for Years 1 and 2 and are positioned to continue development in completion of our original statement of work, or a reasonable modification thereof, in service to DTRA Programs involved in diagnostics and medical countermeasures research and development. The PFP platform is a multi-scale computational modeling system for protein structure-function annotation and function prediction. As of this writing, PFP is the only existing fully automated, high-throughput, multi-scale modeling, whole-proteome annotation platform, and represents a significant advance in the field of genome annotation (Fig. 1). PFP modules perform protein functional annotations at the sequence, systems biology, protein structure, and atomistic levels of biological complexity (Fig. 2). Because these approaches provide orthogonal means of characterizing proteins and suggesting protein function, PFP processing maximizes the protein functional information that can currently be gained by computational means. Comprehensive annotation of pathogen genomes is essential for bio-defense applications in pathogen characterization, threat assessment, and medical countermeasure design and development in that it can short-cut the time and effort required to select and characterize protein biomarkers.

59 BASIC BIOLOGICAL SCIENCES↗

Are Phosphatidic Acids Ubiquitous in Mammalian Tissues or Overemphasized in Mass Spectrometry Imaging Applications?

Abstract Mass spectrometry imaging (MSI) is an invaluable tool for the spatial visualization of molecules in vivo. However, the question of whether observed annotations are endogenous or artificial (i. e., from in‐source fragmentation) is critical and has been largely unexplored in multimodal MSI. In matrix‐assisted laser desorption/ionization (MALDI)‐MSI datasets from researchers worldwide, PAs were found to represent up to 18 % of annotations in rat brain. Rat brain was additionally imaged here using nanospray desorption electrospray ionization (nano‐DESI), a softer ionization strategy. No PAs observed with MALDI were present in the nano‐DESI dataset. Further investigation strongly indicated lipid fragmentation to PAs for MALDI‐MSI, but not with nano‐DESI‐MSI. We finally extend this observation to the MALDI‐MSI analyses of human tissues, showing that PA annotations comprised up to 16 % of annotations. Therefore, this study shows that MSI annotations should be carefully interrogated, as in‐source fragmentation or modification of lipids may contribute substantially to false annotations and incorrect biological interpretations.

Vandergrift, Gregory W.↗

Identification of widespread adenosine nucleotide binding in Mycobacterium tuberculosis

The annotation of protein function is almost completely performed by in silico approaches. However, computational prediction of protein function is frequently incomplete and error prone. In Mycobacterium tuberculosis (Mtb), ~25% of all genes have no predicted function and are annotated as hypothetical proteins. This lack of functional information severely limits our understanding of Mtb pathogenicity. Current tools for experimental functional annotation are limited and often do not scale to entire protein families. Here, we report a generally applicable chemical biology platform to functionally annotate bacterial proteins by combining activity-based protein profiling (ABPP) and quantitative LC-MS-based proteomics. As an example of this approach for high-throughput protein functional validation and discovery, we experimentally annotate the families of ATP-binding proteins in Mtb. Our data experimentally validate prior in silico predictions of >250 ATPases and adenosine nucleotide-binding proteins, and reveal 73 hypothetical proteins as novel ATP-binding proteins. We identify adenosine cofactor interactions with many hypothetical proteins containing a diversity of unrelated sequences, providing a new and expanded view of adenosine nucleotide binding in Mtb. Furthermore, many of these hypothetical proteins are both unique to Mycobacteria and essential for infection, suggesting specialized functions in mycobacterial physiology and pathogenicity. Thus, we provide a generally applicable approach for high throughput protein function discovery and validation, and highlight several ways in which application of activity-based proteomics data can improve the quality of functional annotations to facilitate novel biological insights.

Environmental Molecular Sciences Laboratory↗

Harnessing the predicted maize pan-interactome for putative gene function prediction and prioritization of candidate genes for important traits

Abstract The recent assembly and annotation of the 26 maize nested association mapping population founder inbreds have enabled large-scale pan-genomic comparative studies. These studies have expanded our understanding of agronomically important traits by integrating pan-transcriptomic data with trait-specific gene candidates from previous association mapping results. In contrast to the availability of pan-transcriptomic data, obtaining reliable protein–protein interaction (PPI) data has remained a challenge due to its high cost and complexity. We generated predicted PPI networks for each of the 26 genomes using the established STRING database. The individual genome-interactomes were then integrated to generate core- and pan-interactomes. We deployed the PPI clustering algorithm ClusterONE to identify numerous PPI clusters that were functionally annotated using gene ontology (GO) functional enrichment, demonstrating a diverse range of enriched GO terms across different clusters. Additional cluster annotations were generated by integrating gene coexpression data and gene description annotations, providing additional useful information. We show that the functionally annotated PPI clusters establish a useful framework for protein function prediction and prioritization of candidate genes of interest. Our study not only provides a comprehensive resource of predicted PPI networks for 26 maize genomes but also offers annotated interactome clusters for predicting protein functions and prioritizing gene candidates. The source code for the Python implementation of the analysis workflow and a standalone web application for accessing the analysis results are available at https://github.com/eporetsky/PanPPI.

Genetics & Heredity↗

The Catalytic Site Atlas 2.0: cataloging catalytic sites and residues identified in enzymes

Understanding which are the catalytic residues in an enzyme and what function they perform is crucial to many biology studies, particularly those leading to new therapeutics and enzyme design. The original version of the Catalytic Site Atlas (CSA) (http://www.ebi.ac.uk/thornton-srv/databases/CSA) published in 2004, which catalogs the residues involved in enzyme catalysis in experimentally determined protein structures, had only 177 curated entries and employed a simplistic approach to expanding these annotations to homologous enzyme structures. Here we present a new version of the CSA (CSA 2.0), which greatly expands the number of both curated (968) and automatically annotated catalytic sites in enzyme structures, utilizing a new method for annotation transfer. The curated entries are used, along with the variation in residue type from the sequence comparison, to generate 3D templates of the catalytic sites, which in turn can be used to find catalytic sites in new structures. To ease the transfer of CSA annotations to other resources a new ontology has been developed: the Enzyme Mechanism Ontology, which has permitted the transfer of annotations to Mechanism, Annotation and Classification in Enzymes (MACiE) and UniProt Knowledge Base (UniProtKB) resources. The CSA database schema has been re-designed and both the CSA data and search capabilities are presented in a new modern web interface.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Improve Learning from Crowds via Generative Augmentation

Crowdsourcing provides an efficient label collection schema for supervised machine learning. However, to control annotation cost, each instance in the crowdsourced data is typically annotated by a small number of annotators. This creates a sparsity issue and limits the quality of machine learning models trained on such data. In this paper, we study how to handle sparsity in crowdsourced data using data augmentation. Specifically, we propose to directly learn a classifier by augmenting the raw sparse annotations. We implement two principles of high-quality augmentation using Generative Adversarial Networks: 1) the generated annotations should follow the distribution of authentic ones, which is measured by a discriminator; 2) the generated annotations should have high mutual information with the ground-truth labels, which is measured by an auxiliary network. Extensive experiments and comparisons against an array of state-of-the-art learning from crowds methods on three real-world datasets proved the effectiveness of our data augmentation framework. It shows the potential of our algorithm for low-budget crowdsourcing in general.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Development of an Extensible Computational Framework for Centralized Storage and Distributed Curation and Analysis of Genomic Data Genome-scale Metabolic Models

The DOE funded KBase project of the Stevens group at the University of Chicago was focused on four high-level goals: (i) improve extensibility, accessibility, and scalability of the SEED framework for genome annotation, curation, and analysis; (ii) extend the SEED infrastructure to support transcription regulatory network reconstructions (2.1), metabolic model reconstruction and analysis (2.2), assertions linked to data (2.3), eukaryotic annotation (2.4), and growth phenotype prediction (2.5); (iii) develop a web-API for programmatic remote access to SEED data and services; and (iv) application of all tools to bioenergy-related genomes and organisms. In response to these goals, we enhanced and improved the ModelSEED resource within the SEED to enable new modeling analyses, including improved model reconstruction and phenotype simulation. We also constructed a new website and web-API for the ModelSEED. Further, we constructed a comprehensive web-API for the SEED as a whole. We also made significant strides in building infrastructure in the SEED to support the reconstruction of transcriptional regulatory networks by developing a pipeline to identify sets of consistently expressed genes based on gene expression data. We applied this pipeline to 29 organisms, computing regulons which were subsequently stored in the SEED database and made available on the SEED website (http://pubseed.theseed.org). We developed a new pipeline and database for the use of kmers, or short 8-residue oligomer sequences, to annotate genomes at high speed. Finally, we developed the PlantSEED, or a new pipeline for annotating primary metabolism in plant genomes. All of the work performed within this project formed the early building blocks for the current DOE Knowledgebase system, and the kmer annotation pipeline, plant annotation pipeline, and modeling tools are all still in use in KBase today.

97 MATHEMATICS AND COMPUTING↗

Facility Informatics with GeoVisipedia (Final Report)

Characterization and understanding of nuclear fuel cycle facilities is of considerable interest to DNN R&D and the nonproliferation community due to their importance in nuclear weapons manufacture. Satellite imagery is a key source of information about these facilities, particularly in denied areas. Such imagery can be useful in its own right but annotations applied to the imagery vastly expand the value of the imagery and its usefulness by providing object descriptions, context and potentially linkages to other relevant information. GeoVisipedia (Geospatial Visual Wikipedia) is a new and novel approach to collecting, organizing, sharing and preserving knowledge about nuclear facilities to support nonproliferation efforts. Our approach is unique in that it emphasizes geospatial visual information about nuclear facilities rather than focusing exclusively on textual information. GeoVisipedia users, such as all-source analysts and decision makers, have access to richly annotated nuclear facility imagery with hyperlinks to visual and textual information that provides users with details and context about the facility. A major accomplishment of this project was to develop technology that allows annotations from one satellite image of a facility to be propagated to other satellite imagery of the facility. New satellite imagery of the facility is automatically populated with all previous annotations of the facility. This technology breakthrough provides users with the annotations, and hence knowledge and insight, from all the Subject Matter Experts (SMEs) who have previously annotated imagery of the facility. This capability was demonstrated in a GeoVisipedia prototype, awarded a patent and is described in several publications. In the latter part of the project an informal collaboration was established with the DNN R&D Multi Informatics for Nuclear Operations Scenarios (MINOS) venture. A second prototype of GeoVisipedia was developed that directly supports the MINOS venture. This prototype is called the Nuclear Facility Information Explorer (NFIX) and is tailored to the High Flux Isotope Reactor (HFIR) facility, a testbed for the MINOS venture. NFIX downloads sensor metadata and operator logs, parses and displays this information. NFIX has been licensed and deployed to several laboratories participating in the MINOS venture.

97 MATHEMATICS AND COMPUTING↗

Identification of functional elements and regulatory circuits by Drosophila modENCODE

To gain insight into how genomic information is translated into cellular and developmental programs, the Drosophila model organism Encyclopedia of DNA Elements (modENCODE) project is comprehensively mapping transcripts, histone modifications, chromosomal proteins, transcription factors, replication proteins and intermediates, and nucleosome properties across a developmental time course and in multiple cell lines. We have generated more than 700 data sets and discovered protein-coding, noncoding, RNA regulatory, replication, and chromatin elements, more than tripling the annotated portion of the Drosophila genome. Correlated activity patterns of these elements reveal a functional regulatory network, which predicts putative new functions for genes, reveals stage- and tissue-specific regulators, and enables gene-expression prediction. Our results provide a foundation for directed experimental and computational studies in Drosophila and related species and also a model for systematic data integration toward comprehensive genomic and functional annotation. Several years after the complete genetic sequencing of many species, it is still unclear how to translate genomic information into a functional map of cellular and developmental programs. The Encyclopedia of DNA Elements (ENCODE) (1) and model organism ENCODE (modENCODE) (2) projects use diverse genomic assays to comprehensively annotate the Homo sapiens (human), Drosophila melanogaster (fruit fly), and Caenorhabditis elegans (worm) genomes, through systematic generation and computational integration of functional genomic data sets. Previous genomic studies in flies have made seminal contributions to our understanding of basic biological mechanisms and genome functions, facilitated by genetic, experimental, computational, and manual annotation of the euchromatic and heterochromatic genome (3), small genome size, short life cycle, and a deep knowledge of development, gene function, and chromosome biology. The functions of {approx}40% of the protein and nonprotein-coding genes [FlyBase 5.12 (4)] have been determined from cDNA collections (5, 6), manual curation of gene models (7), gene mutations and comprehensive genome-wide RNA interference screens (8-10), and comparative genomic analyses (11, 12). The Drosophila modENCODE project has generated more than 700 data sets that profile transcripts, histone modifications and physical nucleosome properties, general and specific transcription factors (TFs), and replication programs in cell lines, isolated tissues, and whole organisms across several developmental stages (Fig. 1). Here, we computationally integrate these data sets and report (i) improved and additional genome annotations, including full-length proteincoding genes and peptides as short as 21 amino acids; (ii) noncoding transcripts, including 132 candidate structural RNAs and 1608 nonstructural transcripts; (iii) additional Argonaute (Ago)-associated small RNA genes and pathways, including new microRNAs (miRNAs) encoded within protein-coding exons and endogenous small interfering RNAs (siRNAs) from 3-inch untranslated regions; (iv) chromatin 'states' defined by combinatorial patterns of 18 chromatin marks that are associated with distinct functions and properties; (v) regions of high TF occupancy and replication activity with likely epigenetic regulation; (vi)mixed TF and miRNA regulatory networks with hierarchical structure and enriched feed-forward loops; (vii) coexpression- and co-regulation-based functional annotations for nearly 3000 genes; (viii) stage- and tissue-specific regulators; and (ix) predictive models of gene expression levels and regulator function.

60 APPLIED LIFE SCIENCES↗

Large-Scale Sequencing: The Future of Genomic Sciences Colloquium

Genetic sequencing and the various molecular techniques it has enabled have revolutionized the field of microbiology. Examining and comparing the genetic sequences borne by microbes - including bacteria, archaea, viruses, and microbial eukaryotes - provides researchers insights into the processes microbes carry out, their pathogenic traits, and new ways to use microorganisms in medicine and manufacturing. Until recently, sequencing entire microbial genomes has been laborious and expensive, and the decision to sequence the genome of an organism was made on a case-by-case basis by individual researchers and funding agencies. Now, thanks to new technologies, the cost and effort of sequencing is within reach for even the smallest facilities, and the ability to sequence the genomes of a significant fraction of microbial life may be possible. The availability of numerous microbial genomes will enable unprecedented insights into microbial evolution, function, and physiology. However, the current ad hoc approach to gathering sequence data has resulted in an unbalanced and highly biased sampling of microbial diversity. A well-coordinated, large-scale effort to target the breadth and depth of microbial diversity would result in the greatest impact. The American Academy of Microbiology convened a colloquium to discuss the scientific benefits of engaging in a large-scale, taxonomically-based sequencing project. A group of individuals with expertise in microbiology, genomics, informatics, ecology, and evolution deliberated on the issues inherent in such an effort and generated a set of specific recommendations for how best to proceed. The vast majority of microbes are presently uncultured and, thus, pose significant challenges to such a taxonomically-based approach to sampling genome diversity. However, we have yet to even scratch the surface of the genomic diversity among cultured microbes. A coordinated sequencing effort of cultured organisms is an appropriate place to begin, since not only are their genomes available, but they are also accompanied by data on environment and physiology that can be used to understand the resulting data. As single cell isolation methods improve, there should be a shift toward incorporating uncultured organisms and communities into this effort. Efforts to sequence cultivated isolates should target characterized isolates from culture collections for which biochemical data are available, as well as other cultures of lasting value from personal collections. The genomes of type strains should be among the first targets for sequencing, but creative culture methods, novel cell isolation, and sorting methods would all be helpful in obtaining organisms we have not yet been able to cultivate for sequencing. The data that should be provided for strains targeted for sequencing will depend on the phylogenetic context of the organism and the amount of information available about its nearest relatives. Annotation is an important part of transforming genome sequences into useful resources, but it represents the most significant bottleneck to the field of comparative genomics right now and must be addressed. Furthermore, there is a need for more consistency in both annotation and achieving annotation data. As new annotation tools become available over time, re-annotation of genomes should be implemented, taking advantage of advancements in annotation techniques in order to capitalize on the genome sequences and increase both the societal and scientific benefit of genomics work. Given the proper resources, the knowledge and ability exist to be able to select model systems, some simple, some less so, and dissect them so that we may understand the processes and interactions at work in them. Colloquium participants suggest a five-pronged, coordinated initiative to exhaustively describe six different microbial ecosystems, designed to describe all the gene diversity, across genomes. In this effort, sequencing should be complemented by other experimental data, particularly transcriptomics and metabolomics data, all of which should be gathered and curated continuously. Systematic genomics efforts like the ones outlined in this document would significantly broaden our view of biological diversity and have major effects on science. This has to be backed up with examples. Considering these potential impacts and the need for acquiescence from both the public and scientists to get such projects funded and functioning, education and training will be crucial. New collaborations within the scientific community will also be necessary.

59 BASIC BIOLOGICAL SCIENCES↗

Improvement of eukaryotic protein predictions from soil metagenomes

During the last decades, metagenomics has highlighted the diversity of microorganisms from environmental or host-associated samples. Most metagenomics public repositories use annotation pipelines tailored for prokaryotes regardless of the taxonomic origin of contigs. Consequently, eukaryotic contigs with intrinsically different gene features, are not optimally annotated. Using a bioinformatics pipeline, we have filtered 7.9 billion contigs from 6,872 soil metagenomes in the JGI’s IMG/M database to identify eukaryotic contigs. We have re-annotated genes using eukaryote-tailored methods, yielding 8 million eukaryotic proteins and over 300,000 orphan proteins lacking homology in public databases. Comparing the gene predictions we made with initial JGI ones on the same contigs, we confirmed our pipeline improves eukaryotic proteins completeness and contiguity in soil metagenomes. The improved quality of eukaryotic proteins combined with a more comprehensive assignment method yielded more reliable taxonomic annotation. This dataset of eukaryotic soil proteins with improved completeness, quality and taxonomic annotation reliability is of interest for any scientist aiming at studying the composition, biological functions and gene flux in soil communities involving eukaryotes.

54 ENVIRONMENTAL SCIENCES↗

Scalable and efficient learning from crowds with Gaussian processes

Over the last few years, multiply-annotated data has become a very popular source of information. Online platforms such as Amazon Mechanical Turk have revolutionized the labelling process needed for any classification task, sharing the effort between a number of annotators (instead of the classical single expert). This crowdsourcing approach has introduced new challenging problems, such as handling disagreements on the annotated samples or combining the unknown expertise of the annotators. Probabilistic methods, such as Gaussian Processes (GP), have proven successful to model this new crowdsourcing scenario. However, GPs do not scale up well with the training set size, which makes them prohibitive for medium-to-large datasets (beyond 10K training instances). This constitutes a serious limitation for current real-world applications. In this work, we introduce two scalable and efficient GP-based crowdsourcing methods that allow for processing previously-prohibitive datasets. The first one is an efficient and fast approximation to GP with squared exponential (SE) kernel. The second allows for learning a more flexible kernel at the expense of a heavier training (but still scalable to large datasets). Since the latter is not a GP-SE approximation, it can be also considered as a whole new scalable and efficient crowdsourcing method, useful for any dataset size. Both methods use Fourier features and variational inference, can predict the class of new samples, and estimate the expertise of the involved annotators. A complete experimentation compares them with state-of-the-art probabilistic approaches in synthetic and real crowdsourcing datasets of different sizes. Finally, they stand out as the best performing approach for large scale problems. Moreover, the second method is competitive with the current state-of-the-art for small datasets.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

EXSCLAIM!: Harnessing materials science literature for self-labeled microscopy datasets

This work introduces the EXSCLAIM! toolkit for the automatic extraction, separation, and caption-based natural language annotation of images from scientific literature. EXSCLAIM! is used to show how rule-based natural language processing and image recognition can be leveraged to construct an electron microscopy data set containing thousands of keyword-annotated nanostructure images. Moreover, it is demonstrated how a combination of statistical topic modeling and semantic word similarity comparisons can be used to increase the number and variety of keyword annotations on top of the standard annotations from EXSCLAIM! With large-scale imaging datasets constructed from scientific literature, users are well positioned to train neural networks for classification and recognition tasks specific to microscopy-tasks often otherwise inhibited by a lack of sufficient annotated training data.

36 MATERIALS SCIENCE↗

Rapid Adaptation of Chemical Named Entity Recognition Using Few-Shot Learning and LLM Distillation

Named entity recognition (NER) has been widely used in chemical text mining for the automatic identification and extraction of chemical entities. However, existing chemical NER systems primarily focus on scenarios with abundant training data, requiring significant human effort on annotations. This poses challenges for applications in the chemical field, such as catalysis, where many advancements have traditionally relied on trial-and-error investigations and incremental adjustment of variables. This hinders catalysis science and technology progress in addressing emerging energy and environmental crises. In this work, we propose a few-shot NER model that can quickly adapt to extract new types of chemical entities by using only a limited number of annotated examples. Our model employs a metric-learning approach to transfer entity similarity knowledge from high-resource chemical domains (with abundant annotations) to enable effective entity recognition in low-resource specialized domains (limited annotation). We validate the effectiveness of our model on a few-shot chemical NER benchmark built based on six existing chemical NER data sets. Experiments show that the proposed few-shot NER model can achieve reasonable performance with only 5 examples per entity type and shows consistent improvement as the number of examples increases. Furthermore, we demonstrate how the proposed model can be trained with large language model (LLM) annotated data, opening a new pathway for rapid adaptation of NER systems. Furthermore, our approach leverages the knowledge broadness of large language models for chemistry while distilling this knowledge into a lightweight model suitable for efficient and in-house use.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗