PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Large Language Models”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

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

At least 91 records · Page 5Linked to original sources
PubMed2026

Accelerating inference in genomic and proteomic foundation models via speculative decoding.

MOTIVATION: Genomic and protein foundation models (GFMs and PFMs) have demonstrated strong performance in learning the language of DNA and proteins, but their use in large-scale sequence generation is limited by the latency of autoregressive decoding. Because every token triggers a forward pass of a large Transformer, whose inference is relatively slow, long-sequence generation quickly becomes costly. RESULTS: In this work we adapt speculative decoding to a representative GFM: the DNA model DNAGPT and two representative PFMs: ProGen2 and ProtGPT2. We implement a probabilistic variant of speculative decoding, in which a lightweight draft model proposes short token spans and a larger target model verifies or corrects them in parallel, while preserving the target model's sampling distribution. Across all three models we systematically study the effect of speculation window length, temperature, draft architecture and prompt length, and we benchmark tokens per second over multiple runs per configuration. Speculative decoding yields consistent speedups over standard key-value cached decoding, with maximum observed speedup reaching 100% increase, while average gains across models ranging between 20% and 40% (e.g. 1.2×-1.4×), without changing the underlying target model predictions. Our results show that speculative decoding is a practical and model-agnostic strategy for accelerating genomic and proteomic sequence generation without sacrificing prediction quality. AVAILABILITY AND IMPLEMENTATION: All code and results are freely available at https://github.com/Georgakopoulos-Soares-lab/BioSpecDec.

Genomics↗
PubMed1992

NeuronC: a computational language for investigating functional architecture of neural circuits.

A computational language was developed to simulate neural circuits. A model of a neural circuit with up to 50,000 compartments is constructed from predefined parts of neurons, called "neural elements". A 2-dimensional (2-D) light stimulus and a photoreceptor model allow simulating a visual physiology experiment. Circuit function is computed by integrating difference equations according to standard methods. Large-scale structure in the neural circuit, such as whole neurons, their synaptic connections, and arrays of neurons, are constructed with procedural rules. The language was evaluated with a simulation of the receptive field of a single cone in cat retina, which required a model of cone-horizontal cell network on the order of 1000 neurons. The model was calibrated by adjusting biophysical parameters to match known physiological data. Eliminating specific synaptic connections from the circuit suggested the influence of individual neuron types on the receptive field of a single cone. An advantage of using neural elements in such a model is to simplify the description of a neuron's structure. An advantage of using procedural rules to define connections between neurons is to simplify the network definition.

Computer Simulation↗
PubMed2026

CodonMoE: DNA language models for codon-dependent mRNA prediction.

MOTIVATION: Genomic language models (gLMs) face a fundamental efficiency challenge: one must either maintain separate specialized models for each biological modality (DNA and RNA) or develop large multimodal architectures. Both approaches impose significant computational burdens-modality-specific models require redundant infrastructure despite inherent biological connections, while multi-modal architectures demand increased parameter counts and extensive cross-modality pretraining. RESULTS: To address this limitation, we introduce CodonMoE (Adaptive Mixture of Codon Reformative Experts), a lightweight adapter that transforms DNA language models into effective RNA analyzers without RNA-specific pretraining. Our theoretical analysis establishes CodonMoE as a universal approximator at the codon level, capable of mapping arbitrary functions from codon sequences to codon-dependent RNA properties given sufficient expert capacity. Across four RNA prediction tasks spanning stability, expression, and regulation, DNA models augmented with CodonMoE significantly outperform their unmodified counterparts, with the HyenaDNA+CodonMoE series achieving state-of-the-art results using 80% fewer parameters than specialized RNA models. By maintaining sub-quadratic complexity while achieving superior performance, our approach provides a principled path toward unifying genomic language modeling, leveraging more abundant DNA data and reducing computational overhead while preserving modality-specific performance advantages. AVAILABILITY AND IMPLEMENTATION: Source code for the method and to reproduce the results is available at https://github.com/Kingsford-Group/CodonMoE.

Codon↗
PubMed2026

CeLLTra: aligning cell names with gene expression via a pathway-informed transformer.

MOTIVATION: Single-cell RNA sequencing (scRNA-Seq) technology enables detailed exploration of gene expression at the individual cell level, crucial for annotating cell types and understanding cellular diversity. Traditional methods for cell type annotation often rely on marker genes and manual labeling, posing challenges due to low data quality and incomplete reference datasets. RESULTS: We developed CeLLTra, a novel contrastive learning framework that leverages a Transformer-based model integrating biological pathway information to group genes into super tokens, effectively capturing comprehensive gene expression from scRNA-Seq data. By combining this pathway-informed Transformer with a pretrained domain-specific language model, CeLLTra accurately aligns cell-type annotations with gene expression profiles. Evaluations on a large-scale human scRNA-Seq dataset showed that CeLLTra significantly outperformed state-of-the-art methods in supervised and zero-shot cell-type prediction. Additionally, CeLLTra generalized well to external datasets, improving clustering performance and enabling better characterization of cancerous cell states in tumor-infiltrating myeloid cells from non-small cell lung cancer patients. AVAILABILITY AND IMPLEMENTATION: CeLLTra is freely available on GitHub (https://github.com/WJZheng-group/CeLLTra) and Zenodo (https://doi.org/10.5281/zenodo.17666735). The datasets underlying this article are the following: GSE201333 and GSE127465. All these datasets are publicly available and can be freely accessed on the Gene Expression Omnibus repository.

Humans↗
PubMedDate not supplied

A Canadian perspective on learning disabilities.

Canadian practice and research with children and adults with learning disabilities are described and analyzed. After an examination of the historical basis for current practice, the societal and cultural factors affecting education of children with learning disabilities, services for adults, and research are discussed. It was found that policy and legislation regarding special education vary considerably from province to province, and identification practices and service delivery models vary even within provinces. The fact that Canada has two official languages (English and French), a large multicultural community, and a Native population with special needs often arising from poverty has an impact on the education of children with learning disabilities and on sample description in research. Although school-age children are relatively well served, services for preschool children and adults with learning disabilities are minimal. The positive features of Canadian service delivery are that most programs are publicly funded, decision making tends to be nonadversarial and collaborative, and the needs of the whole child are typically considered.

Adult↗
PubMed1993

Observations on the recent history of drug user counseling.

The drug user counselor role is explored in terms of its changing nature over the course of the past 25 years. Initially, the drug user counselor could be characterized as a professional based on his or her experience, or as an ex-addict paraprofessional in the language of that time. Working very largely with a heroin-using clientele, the counselor was the advisor and role model who could not be conned and, thereby, the essential counterpart to the mental health professionals who were entering the drug misuse field. Over time, these latter professionals based on education have become increasingly evident and the "professionals of experience" have become less so in accord with changes in the demography, drug-using characteristics, and psychological functioning of drug user clients. Nonetheless, studies that support the particular efficacy of counselors of education for all but drug user clients with significant psychopathology are lacking. Moreover, aspects of therapeutic interaction that are more largely engaged in by "professionals of experience" are threatened by the diminution in that group's numbers and the credentialing out of nontraditional job functions. Over the past few years, awareness of the significance of the contributions of "professionals of experience" has been reawakened by the threat of AIDS and the recognition of counselors' contributions to outreach and AIDS prevention counseling.

Acquired Immunodeficiency Syndrome↗
PubMed2025

Leveraging protein language models for cross-variant CRISPR/Cas9 sgRNA activity prediction.

MOTIVATION: Accurate prediction of single-guide RNA (sgRNA) activity is crucial for optimizing the CRISPR/Cas9 gene-editing system, as it directly influences the efficiency and accuracy of genome modifications. However, existing prediction methods mainly rely on large-scale experimental data of a single Cas9 variant to construct Cas9 protein (variants)-specific sgRNA activity prediction models, which limits their generalization ability and prediction performance across different Cas9 protein (variants), as well as their scalability to the continuously discovered new variants. RESULTS: In this study, we proposed PLM-CRISPR, a novel deep learning-based model that leverages protein language models to capture Cas9 protein (variants) representations for cross-variant sgRNA activity prediction. PLM-CRISPR uses tailored feature extraction modules for both sgRNA and protein sequences, incorporating a cross-variant training strategy and a dynamic feature fusion mechanism to effectively model their interactions. Extensive experiments demonstrate that PLM-CRISPR outperforms existing methods across datasets spanning seven Cas9 protein (variants) in three real-world scenarios, demonstrating its superior performance in handling data-scarce situations, including cases with few or no samples for novel variants. Comparative analyses with traditional machine learning and deep learning models further confirm the effectiveness of PLM-CRISPR. Additionally, motif analysis reveals that PLM-CRISPR accurately identifies high-activity sgRNA sequence patterns across diverse Cas9 protein (variants). Overall, PLM-CRISPR provides a robust, scalable, and generalizable solution for sgRNA activity prediction across diverse Cas9 protein (variants). AVAILABILITY AND IMPLEMENTATION: The source code can be obtained from https://github.com/CSUBioGroup/PLM-CRISPR.

CRISPR-Cas Systems↗
PubMed1994

A nonlinear electrical-thermal model of the skin.

This work presents a model for the skin which accounts for both the nonlinearities and the asymmetries in its voltage-current characteristic. This model consists of an electrical submodel and a heat transfer submodel. The electrical submodel uses nonlinear devices in which some parameters depend on skin temperature. The heat transfer submodel models the heat exchange between the skin, the surrounding tissues, and the ambient medium and calculates the temperature of the skin to update the necessary parameters of the electrical submodel. The model is based on experiments designed to determine: 1) the dry skin voltage-current characteristic; 2) the changes in the skin breakdown voltage with location; 3) the moist skin voltage-current characteristic; 4) the changes in the voltage-current characteristic of the skin with duration after the onset of stimulation; and 5) the effect of skin temperature on its voltage-current characteristic. During these experiments we used 84-mm2 square Ag-AgCl electrodes to apply sinusoidal voltage of 0.2 and 20 Hz. The simulations were performed using the Advanced Continuous Simulation Language (ACSL), capable of solving differential and integral equations with variable coefficients. The model predicted the skin behavior satisfactorily for a large range of amplitudes and frequencies. We found that the breakdown occurred when the energy delivered to the skin exceeded a threshold. Above this threshold the voltage-current characteristic of the skin became nonlinear and asymmetric and, in a real situation, the subject would experience an uncomfortable sensation which could rapidly develop into pain.

Electric Conductivity↗
PubMedDate not supplied

Medical image databases: a content-based retrieval approach.

Information contained in medical images differs considerably from that residing in alphanumeric format. The difference can be attributed to four characteristics: (1) the semantics of medical knowledge extractable from images is imprecise; (2) image information contains form and spatial data, which are not expressible in conventional language; (3) a large part of image information is geometric; (4) diagnostic inferences derived from images rest on an incomplete, continuously evolving model of normality. This paper explores the differentiating characteristics of text versus images and their impact on design of a medical image database intended to allow content-based indexing and retrieval. One strategy for implementing medical image databases is presented, which employs object-oriented iconic queries, semantics by association with prototypes, and a generic schema.

Abstracting and Indexing↗
PubMedDate not supplied

A commercial large-vocabulary discrete speech recognition system: DragonDictate.

DragonDictate is currently the only commercially available general-purpose, large-vocabulary speech recognition system. It uses discrete speech and is speaker-dependent, adapting to the speaker's voice and language model with every word. Its acoustic adaptability is based in a three-level phonology and a stochastic model of production. The phonological levels are phonemes, augmented triphones (phonemes-in-context or PICs), and steady-state spectral slices that are concatenated to approximate the spectra of these PICs (phonetic elements or PELs) and thus of words. Production is treated as a hidden Markov process, which the recognizer has to identify from its output, the spoken word. Findings of practical value to speech recognition are presented from research on six European languages.

Female↗
PubMed1994

Generating patient-specific interactive natural language explanations.

Patient compliance is a significant problem and is strongly correlated with the patients' understanding of their condition and prescribed treatment. Since doctors typically do not have large amounts of time to educate patients, and impersonal, voluminous patient handouts are largely ineffective, we propose the use of a sophisticated computer-based information system to generate tailored, interactive handouts to communicate with patients. Our system uses text planning and user modeling techniques to generate natural language descriptions of migraine, its symptoms, triggering factors and prescriptions. The system is capable of handling follow-up questions requesting further information, and generating responses in the context of previously supplied information--a capability unavailable in previous patient information systems. The system tailors its interaction to: (i) the class of migraine patients, (ii) the individual patient, and (iii) the previous dialogue. Preliminary evaluation of the system indicates that patients find it useful and informative. More extensive evaluation is in progress.

Computer-Assisted Instruction↗
PubMed1995

Indo-European origins: a computer-simulation test of five hypotheses.

Allele frequency distributions were generated by computer simulation of five models of microevolution in European populations. Genetic distances calculated from these distributions were compared with observed genetic distances among Indo-European speakers. The simulated models differ in complexity, but all incorporate random genetic drift and short-range gene flow (isolation by distance). The best correlations between observed and simulated data were obtained for two models where dispersal of Neolithic farmers from the Near East depends only on population growth. More complex models, where the timing of the farmers' expansion is constrained by archaeological time data, fail to account for a larger fraction of the observed genetic variation; this is also the case for a model including late Neolithic migrations from the Pontic steppes. The genetic structure of current populations speaking Indo-European languages seems therefore to largely reflect a Neolithic expansion. This is consistent with the hypothesis of a parallel spread of farming technologies and a proto-Indo-European language in the Neolithic. Allele-frequency gradients among Indo-European speakers may be due either to incomplete admixture between dispersing farmers, who presumably spoke proto-Indo-European, and pre-existing hunters and gatherers (as in the traditional demic diffusion hypothesis), or to founder effects during the farmers' dispersal. By contrast, successive migrational waves from the East, if any, do not seem to have had genetic consequences detectable by the present comparison of observed and simulated allele frequencies.

Alleles↗
PubMed1997

Content-based indexing of images and video.

By representing image content using probabilistic models of an object's appearance we can obtain semantics-preserving compression of the image data. Such compact representations of an image's salient features allow rapid computer searches of even large image databases. Examples are shown for databases of face images, a video of American sign language (ASL), and a video of facial expressions.

Algorithms↗
PubMed1988

Sequentiality of speech acts in conversational structure.

One aspect of the phenomenon of coherence in conversational discourse was addressed in the present study: sequentiality of speech acts. Several models of discourse structure have postulated sequencing rules between speech acts in conversations, but these efforts have been hampered by the lack of an efficient empirical method that can characterize a large body of language data. The lag sequential technique is proposed here as a tool that can be used to abstract a "grammar" of speech act contingency from spoken discourse. Derived patterns of discourse between female adults and preschool children confirmed expectations that most discourse is based upon three fundamental speech act pairings: question--answer, statement--reply, and directive--acknowledgement. It was also found that interlocutor differences in status, knowledge, and conversational ability affected the structure of the discourse in predictable ways.

Child↗
PubMedDate not supplied

Age, cohort, and time development muddles: easy in practice, hard in theory.

The debate among developmental psychologists over how best to combine longitudinal and cross-sectional data sequences can be traced back at least four decades. During the 1970s, a variation of this theme received much attention: could the developmental influences of age, cohort, and time be unraveled by sufficiently ingenious application of combined data sequences? We believe discussion of this question has been needlessly parochial and confused. Substantive and methodological contributions from other disciplines have, until recently, been largely ignored by developmental psychologists. Moreover, the solutions debated by psychologists have generally been formulated in language that obscured, rather than explicated, the formal indeterminacy implicit in models of age, cohort, and time parameters. Methodologists have outlined formal solutions to problems of indeterminacy in the contexts of model identification and estimability theory. Sociologists have proposed solutions, based on explicitly theoretical assumptions, that permit model identification and unambiguous interpretation. We review these contributions, and suggest a hierarchy of solutions to the problem of age, cohort, and time indeterminacy.

Aging↗
PubMed1993

A conceptual model for information retrieval with UMLS.

Information retrieval in large information databases is a non-deterministic process which needs a sequence of search steps generally. One of the main problems to which the end-users are faced is to parse efficiently their questions into the query language that the computer systems allow. Conceptual graphs were initially designed for natural language analysis and understanding. Due to their closeness to semantic networks, their expressiveness is powerful enough to be applied to knowledge representation and use by computer systems. This work demonstrates that conceptual graphs are a suitable means to model the end-users querieson the basis of the thesaurus and the semantic network of the UMLS project.

Information Storage and Retrieval↗
PubMed2025

Design of highly functional genome editors by modelling CRISPR-Cas sequences.

Gene editing has the potential to solve fundamental challenges in agriculture, biotechnology and human health. CRISPR-based gene editors derived from microorganisms, although powerful, often show notable functional tradeoffs when ported into non-native environments, such as human cells1. Artificial-intelligence-enabled design provides a powerful alternative with the potential to bypass evolutionary constraints and generate editors with optimal properties. Here, using large language models2 trained on biological diversity at scale, we demonstrate successful precision editing of the human genome with a programmable gene editor designed with artificial intelligence. To achieve this goal, we curated a dataset of more than 1 million CRISPR operons through systematic mining of 26 terabases of assembled genomes and metagenomes. We demonstrate the capacity of our models by generating 4.8× the number of protein clusters across CRISPR-Cas families found in nature and tailoring single-guide RNA sequences for Cas9-like effector proteins. Several of the generated gene editors show comparable or improved activity and specificity relative to SpCas9, the prototypical gene editing effector, while being 400 mutations away in sequence. Finally, we demonstrate that an artificial-intelligence-generated gene editor, denoted as OpenCRISPR-1, exhibits compatibility with base editing. We release OpenCRISPR-1 to facilitate broad, ethical use across research and commercial applications.

CRISPR-Cas Systems↗