SEARCH · PubMed Health

Results for “large language model”

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 55 records · Page 3Linked to original sources

PubMed2025

Embed-Search-Align: DNA sequence alignment using Transformer models.

MOTIVATION: DNA sequence alignment, an important genomic task, involves assigning short DNA reads to the most probable locations on an extensive reference genome. Conventional methods tackle this challenge in two steps: genome indexing followed by efficient search to locate likely positions for given reads. Building on the success of Large Language Models in encoding text into embeddings, where the distance metric captures semantic similarity, recent efforts have encoded DNA sequences into vectors using Transformers and have shown promising results in tasks involving classification of short DNA sequences. Performance at sequence classification tasks does not, however, guarantee sequence alignment, where it is necessary to conduct a genome-wide search to align every read successfully, a significantly longer-range task by comparison. RESULTS: We bridge this gap by developing a "Embed-Search-Align" (ESA) framework, where a novel Reference-Free DNA Embedding (RDE) Transformer model generates vector embeddings of reads and fragments of the reference in a shared vector space; read-fragment distance metric is then used as a surrogate for sequence similarity. ESA introduces: (i) Contrastive loss for self-supervised training of DNA sequence representations, facilitating rich reference-free, sequence-level embeddings, and (ii) a DNA vector store to enable search across fragments on a global scale. RDE is 99% accurate when aligning 250-length reads onto a human reference genome of 3 gigabases (single-haploid), rivaling conventional algorithmic sequence alignment methods such as Bowtie and BWA-Mem. RDE far exceeds the performance of six recent DNA-Transformer model baselines such as Nucleotide Transformer, Hyena-DNA, and shows task transfer across chromosomes and species. AVAILABILITY AND IMPLEMENTATION: Please see https://anonymous.4open.science/r/dna2vec-7E4E/readme.md.

Sequence Analysis, DNA↗

PubMed2026

Multimodal alignment improves generalizability of genomic biomarker prediction in computational pathology.

Computational pathology models that use digitized histopathology whole-slide images have the potential to become a cost-effective and scalable alternative to molecular assays for the prediction of genomic biomarkers, a key task in precision oncology. However, as new genomic biomarkers are discovered or quantified, large, labeled datasets must be prospectively collected to train new models. To address this challenge, we developed multimodal alignment for biomarker learning and generalization (MARBLE), a multimodal contrastive pretraining strategy that integrates structured biomarker knowledge into representation learning of histopathology images. MARBLE aligns histopathology-derived representations with representations of genomic biomarkers generated by a large language model (LLM) and a protein language model (PLM). This biologically informed alignment enables data-efficient generalization to novel, out-of-distribution biomarkers. Using the MSK-IMPACT cohort of over 40,000 patients across multiple biomarker panel versions, we design experiments grounded in real-world data to demonstrate the value of our proposed approach.

CP: computational biology↗

PubMed2026

The Continuity Trap in Data Science Health Research.

Secondary use is now the ordinary condition of data science health research rather than an exception to it. Electronic health records collected for clinical care become prediction tools and inputs for generative AI; imaging archives become foundation-model corpora; genomic datasets become resources for polygenic risk scores; and legacy biospecimens become renewable, indefinitely distributable cell lines. Governance has responded by emphasizing verifiable instruments such as provenance logs, repository approvals, broad-consent forms, data-use agreements, model cards, records of processing, and locality-preserving architectures. These instruments are necessary, and they answer real questions about lineage, privacy, institutional responsibility, and accountability, but they are not sufficient to establish that a present use remains ethically justified. We define ethical continuity as the persistence of normatively relevant relationships between the original conditions of data generation or material collection and subsequent downstream uses, such that current uses remain justifiable in light of the expectations, permissions, meanings, and relational obligations present at entrustment. We then define the Continuity Trap as a review-stage governance error in which a salient signal of continuity in one domain is treated as sufficient evidence of ethical continuity overall, causing inquiry into the remaining domains to close prematurely. The trap is not ordinary noncompliance, ethics creep, or a demand for universal rereview; it is a cross-domain inference error that can arise even in careful, good-faith review. We distinguish it from proxy closure, of which it is a continuity-specific subtype, and from Goodhart's and Campbell's laws, which describe how measures degrade once they become targets. We operationalize ethical continuity across 4 domains: provenance, semantics, authorization, and relational standing, developed in our Representational Veracity framework, and we show that these domains can diverge as data are linked, transformed, modeled, and redeployed. We identify the institutional mechanisms-provenance privilege, descriptor sedimentation, authorization fossilization, and community effacement-that cause auditable signals to be overread, and we examine how the US Health Insurance Portability and Accountability Act (HIPAA) of 1996, the General Data Protection Regulation, the European Health Data Space, US Food and Drug Administration guidance, the US National Institute of Standards and Technology (NIST) AI Risk Management Framework, and federated-learning governance can reduce risk while still inducing continuity traps. We apply the framework to consent and nonconsent settings, including public health, immunization, syndromic, and wastewater surveillance, polygenic risk scores, induced pluripotent stem cells, federated learning, and health-related large language models. The policy implication is trigger-based continuity review: rather than rereviewing every reuse, investigators and reviewers should identify the weakest continuity domain at the present data stage and impose a domain-matched safeguard, recorded in a short continuity statement. This reframing is intended for the committees, repositories, funders, and governance bodies that decide whether reuse may proceed, and it matters most in cross-border and low-resource settings. Provenance should begin ethical review; it should not end it.

Data Science↗

PubMed2025

Identification and classification of ion-channels across the tree of life provide functional insights into understudied CALHM channels.

The ion channel (IC) genes encoded in the human genome play fundamental roles in cellular functions and disease and are one of the largest classes of druggable proteins. However, limited knowledge of the diverse molecular and cellular functions carried out by ICs presents a major bottleneck in developing selective chemical probes for modulating their functions in disease states. The wealth of sequence data available on ICs from diverse organisms provides a valuable source of untapped information for illuminating the unique modes of channel regulation and functional specialization. However, the extensive diversification of IC sequences and the lack of a unified resource present a challenge in effectively using existing data for IC research. Here, we perform integrative mining of available sequence, structure, and functional data on 419 human ICs across disparate sources, including extensive literature mining by leveraging advances in large language models to annotate and curate the full complement of the "channelome". We employ a well-established orthology inference approach to identify and extend the IC orthologs across diverse organisms to above 48,000. We show that the depth of conservation and taxonomic representation of IC sequences can further be translated to functional similarities by clustering them into functionally relevant groups, which can be used for downstream functional prediction on understudied members. We demonstrate this by delineating co-conserved patterns characteristic of the understudied family of the Calcium Homeostasis Modulator (CALHM) family of ICs. Through mutational analysis of co-conserved residues altered in human diseases and electrophysiological studies, we show that these evolutionarily-constrained residues play an important role in channel gating functions. Thus, by providing new tools and resources for performing large comparative analyses on ICs, this study addresses the unique needs of the IC community and provides the groundwork for accelerating the functional characterization of dark channels for therapeutic intervention.

CALHM1↗

PubMed2025

AskBeacon-performing genomic data exchange and analytics with natural language.

MOTIVATION: Enabling clinicians and researchers to directly interact with global genomic data resources by removing technological barriers is vital for medical genomics. AskBeacon enables large language models (LLMs) to be applied to securely shared cohorts via the Global Alliance for Genomics and Health Beacon protocol. By simply "asking" Beacon, actionable insights can be gained, analyzed, and made publication-ready. RESULTS: In the Parkinson's Progression Markers Initiative (PPMI), we use natural language to ask whether the sex-differences observed in Parkinson's disease are due to X-linked or autosomal markers. AskBeacon returns a publication-ready visualization showing that for PPMI the autosomal marker occurred 1.4 times more often in males with Parkinson's disease than females, compared to no differences for the X-linked marker. We evaluate commercial and open-weight LLM models, as well as different architectures to identify the best strategy for translating research questions to Beacon queries. AskBeacon implements extensive safety guardrails to ensure that genomic data is not exposed to the LLM directly, and that generated code for data extraction, analysis and visualization process is sanitized and hallucination resistant, so data cannot be leaked or falsified. AVAILABILITY AND IMPLEMENTATION: AskBeacon is available at https://github.com/aehrc/AskBeacon.

Genomics↗

PubMed2025

AI-HOPE: an AI-driven conversational agent for enhanced clinical and genomic data integration in precision medicine research.

MOTIVATION: The growing complexity of clinical cancer research has fueled a surge in demand for automated bioinformatics tools capable of integrating clinical and genomic data to accelerate discovery efforts. RESULTS: We present the Artificial Intelligence Agent for High-Optimization and Precision Medicine (AI-HOPE), an AI-driven system that enables domain experts to conduct integrative data analyses through natural language interactions. Powered by Large Language Models, AI-HOPE interprets user instructions, converts them into executable code, and autonomously analyzes locally stored data. It supports flexible association studies, subset comparisons, clinical prevalence assessments and survival analyses. In addition, AI-HOPE enables global variable scans to identify features significantly associated with a user-defined outcome, making a powerful and intuitive tool for advancing precision medicine research. Importantly, its closed-system design prevents clinical data leakage. To demonstrate its utility, AI-HOPE was applied to The Cancer Genome Atlas data to address two clinical questions. First, it identified significant enrichment of TP53 mutations in late-stage colorectal cancer compared to early-stage cases. Second, it uncovered a strong association between KRAS mutations and poorer progression-free survival in FOLFOX-treated patients. These findings align with established literature and demonstrate AI-HOPE's ability to generate meaningful insights independently, without prior assumptions. By removing programming barriers and simplifying complex analyses, AI-HOPE bridges the gap between data complexity and research needs. With its scalable and adaptable framework, AI-HOPE has the potential to support diverse biomedical research fields, driving innovation and efficiency in translational studies. AVAILABILITY AND IMPLEMENTATION: The AI-HOPE software and demonstration data is available at https://github.com/Velazquez-Villarreal-Lab/AI-HOPE.

Precision Medicine↗

PubMed2025

AutoPM3: enhancing variant interpretation via LLM-driven PM3 evidence extraction from scientific literature.

MOTIVATION: Rare diseases affect over 300 million people worldwide and are often caused by genetic variants. While variant detection has become cost-effective, interpreting these variants-particularly collecting literature-based evidence like ACMG/AMP PM3-remains complex and time-consuming. RESULTS: We present AutoPM3, a method that automates PM3 evidence extraction from literatures using open-source large language models (LLMs). AutoPM3 combines a Text2SQL-based variant extractor and a retrieval-augmented generation (RAG) module, enhanced by a variant-specific retriever and fine-tuned LLM, to separately process tables and text. We curated PM3-Bench, a dataset of 1027 variant-publication evidence pairs from ClinGen. On openly accessible pairs, AutoPM3 achieved 86.1% accuracy for variant hits and 72.5% recall for in trans variants-outperforming other methods, including those using larger models. We uncovered the effectiveness of AutoPM3's key modules, especially for variant-specific retriever and Text2SQL, through the sequential ablation study. AutoPM3 located evidence in 76 s, demonstrating that open-source LLMs can offer an efficient, cost-effective solution for rare disease diagnosis. AVAILABILITY AND IMPLEMENTATION: AutoPM3 is implemented and freely available under the MIT license at https://github.com/HKU-BAL/AutoPM3.

Genetic Variation↗

PubMed2025

Pre-Meta: priors-augmented retrieval for LLM-based metadata generation.

MOTIVATION: While high-throughput sequencing technologies have dramatically accelerated genomic data generation, the manual processes required for dataset annotation and metadata creation impede the efficient discovery and publication of these resources across disparate public repositories. Large language models (LLMs) have the potential to streamline dataset profiling and discovery. However, their current limitations in generalizing across specialized knowledge domains, particularly in fields such as biomedical genomics, prevent them from fully realizing this potential. This article presents Pre-Meta, an LLM-agnostic and domain-independent data annotation pipeline with an enriched retrieval procedure that leverages related priors-such as pre-generated metadata tags and ontologies-as auxiliary information to improve the accuracy of automated metadata generation. RESULTS: Validated using five selected metadata fields sampled across 1500 papers, the Pre-Meta assisted annotation experiment-without finetuning and prompt optimization-demonstrates a systemic improvement in the annotation task: shown through a 23%, 72%, and 75% accuracy gain from conventional RAG adoptions of GPT-4o mini, Llama 8B, and Mistral 7B respectively. AVAILABILITY AND IMPLEMENTATION: The code, data access, and scripts are available at: https://github.com/SINTEF-SE/LLMDap.

Metadata↗

PubMed2025

Target and biomarker exploration portal for drug discovery.

MOTIVATION: The discovery of novel drug targets and precision biomarkers remains a major challenge in drug development, with traditional differential expression analysis often overlooking key regulatory proteins. Here, we present a novel, web-based bioinformatics tool, the Target and Biomarker Exploration Portal (TBEP), designed to accelerate the drug discovery process by integrating large-scale biomedical data with network analysis techniques. RESULTS: TBEP harnesses machine-learning approaches to mine and combine multimodal datasets, including human genetics, functional genomics, and protein-protein interaction networks, to decode causal disease mechanisms and uncover novel therapeutic targets and precision biomarkers for specific phenotypes. A unique feature of the tool is its ability to process large-scale data in real-time, facilitated by an efficient cloud-based architecture. Additionally, the tool incorporates an integrated large language model (LLM), which assists researchers in exploring and interpreting complex biological relationships within the generated networks and multi-omics data using natural language (English). By offering an intuitive, interactive interface, the LLM enhances the exploration of biological insights, making it easier for scientists to derive actionable conclusions. This powerful integration of network analysis, multi-omics data, and LLM provides a robust framework for accelerating the identification of novel drug targets. AVAILABILITY AND IMPLEMENTATION: The tool is publicly available at https://tbep.missouri.edu. The source code, documentation and installation instructions are available at GitHub repository: https://github.com/mizzoudbl/tbep.

Drug Discovery↗

PubMed2026

BioMedGraphica: an all-in-one platform for joint textual biomedical prior knowledge and numeric graph generation.

MOTIVATION: Multiomics data analysis is essential for scientific discovery in precision medicine. However, translating analysis results of omics data analysis into novel scientific hypotheses remains a significant challenge. Human experts must manually review analysis results and generate new hypotheses based on extensive and interconnected biomedical prior knowledge, which is subjective and not scalable. While large language models can accelerate the discovery, their reasoning improves when grounded in structured, auditable, and comprehensive biomedical prior knowledge. However, biomedical knowledge is scattered across heterogeneous databases that use diverse and inconsistent nomenclature systems, making it difficult to integrate resources into a unified format for scalable analysis. This fragmentation limits the ability of artificial intelligence systems to fully leverage biomedical data for scientific discovery. RESULTS: We developed BioMedGraphica, a novel all-in-one platform that harmonizes fragmented biomedical resources by integrating 11 entity types and 30 relation types from 43 databases into a unified textual prior knowledge graph containing 2 306 921 entities and 27 232 091 relations. In addition, we present a novel textual-numeric graph (TNG) data structure concept, where textual information captures prior biological knowledge (e.g. transcription start sites, functions, mechanisms), numeric values represent quantitative biomedical features, and the integrated relations can help uncover mechanisms. By bridging prior knowledge with user-specific data, TNG is a novel and ideal data structure for developing novel graph analysis models. AVAILABILITY AND IMPLEMENTATION: The code is available at: https://github.com/FuhaiLiAiLab/BioMedGraphica and BioMedGraphica knowledge graph database can be downloaded from huggingface dataset: https://huggingface.co/datasets/FuhaiLiAiLab/BioMedGraphica.

Humans↗

PubMed2025

BioMedGraphica: An All-in-One Platform for Joint Textual Biomedical Prior Knowledge and Numeric Graph Generation.

Multi-omic data analysis is essential for scientific discovery in precision medicine. However, translating statistical results of omic data analysis into novel scientific hypothesis remains a significant challenge. Human experts must manually review analysis results and generate new hypothesis based on extensive and inter-connected biomedical prior knowledge, which is subjective and not scalable. While large language models (LLMs) can accelerate the discovery, their reasoning improves when grounded in structured, auditable and comprehensive biomedical prior knowledge. Biomedical knowledge, however, is scattered across heterogeneous databases that use diverse and inconsistent nomenclature systems, making it difficult to integrate resources into a unified format for scalable analysis. This fragmentation limits the ability of AI systems to fully leverage biomedical data for scientific discovery. To address these challenges, we developed BioMedGraphica , an all-in-one platform that harmonizes fragmented biomedical resources by integrating 11 entity types and 30 relation types from 43 databases into a unified knowledge graph containing 2,306,921 entities and 27,232,091 relations. In addition, to the best of our knowledge, this is the first work to propose a novel Textual-Numeric Graph (TNG) data-structure for multi-omics data analysis. In TNG, textual information captures prior biological knowledge (e.g., transcription start sites, functions, mechanisms), while numeric values represent quantitative biomedical features, and the integrated relations can help uncover mechanisms. By bridging prior knowledge with user-specific data, TNG is a novel and ideal data-structure for the development of graph foundation models, with the potential to improve prediction performance and interpretability, while also augmenting LLMs by supplying graph-structured mechanistic context to strengthen reasoning. The details for BioMedGraphica code can be accessed by github link: https://github.com/FuhaiLiAiLab/BioMedGraphica and BioMedGraphica knowledge graph data can be downloaded from huggingface dataset: https://huggingface.co/datasets/FuhaiLiAiLab/BioMedGraphica.

biomedical knowledge graph↗

PubMed2026

A trainable language model with potential to modulate translation rates in non-model organisms by generating upstream untranslated region sequence libraries.

Tuning protein expression in non-model organisms is often constrained by the lack of validated genetic parts and predictive design tools. Translational tuning through the modulation of upstream untranslated regions (5'-UTRs) offers a potentially organism-agnostic route, but existing methods typically rely on mechanistic assumptions, prior knowledge that may not be available in non-model contexts, or the screening of sequence libraries. Here, we present a simple generative approach for creating synthetic 5'-UTR libraries based solely on the genomic sequence statistics of any desired organism. The method uses a sliding-window n-gram language model applied to native 5'-UTR sequences to produce novel sequences that preserve organism-specific base distributions and motifs without hard-coding specific motifs or mechanistic rules into inflexible statistical templates. We have applied this approach to the model bacterium Escherichia coli and the non-model probiotic Limosilactobacillus reuteri. Libraries of approximately 1,000 sequences were generated for each organism, from which about 100 unique sequences were experimentally tested for translation of a fluorescent reporter protein. In both organisms, the synthetic libraries yielded a broad range of translation levels from this relatively small number of tested variants. Sequences derived from an organism's own genomic statistics provided a more uniformly distributed range of translation rates in that organism than sequences derived from the other species. Correlations of individual sequence performance across the two species were weak, and thermodynamic predictions of ribosome binding strength showed very little predictive power, especially in the non-model L. reuteri. The results demonstrate that simple statistical language model approaches applied to genomic data can generate functional translational regulatory sequence libraries without detailed mechanistic knowledge or explicit reference to consensus motifs. The approach requires minimal computational resources, avoids reproducing native sequences, and can be readily applied to any organism with a sequenced genome. This strategy may lower technical barriers to expression tuning in non-model organisms.

5' Untranslated Regions↗

PubMed2026

Investigating cross-organism prediction of prokaryotic essential proteins using unsupervised language model and ensemble strategy.

Cross-organism prediction of essential proteins is a critical task for drug discovery and microbial engineering, yet the generalizability of existing machine learning models across diverse species remains a significant challenge. In this study, we propose DeepPEP, a large language model-based framework designed to reliably transfer essential protein annotations between distantly related organisms. Utilizing 66 curated prokaryotic datasets, we systematically evaluated DeepPEP's cross-organism performance under various conditions. Initial pairwise predictions revealed a correlation between performance and evolutionary distance; however, further investigation demonstrated that integrating training data from multiple organisms yields superior predictive power. In a benchmark scenario designed to simulate real-world applications, DeepPEP outperformed the state-of-the-art tool Geptop 2.0, showcasing a robust ability to identify species-specific essential proteins. Finally, a case study on novel genomes confirmed the model's practical effectiveness. Our results suggest that DeepPEP is a powerful strategy for prokaryotic essential protein prediction, and the rigorous evaluation framework established in this study provides a new benchmark for the field.

Large Language Models↗

PubMed2026

An open benchmark and language models for AI in aging biology.

Over the past two decades, human aging has been characterized across DNA methylation, transcriptomic, proteomic, and clinical modalities, yet no benchmark evaluates whether AI systems can interpret these heterogeneous data types in the context of aging biology. We introduce LongevityBench, an open suite of 17 tasks spanning five biodata domains, and use it to assess 18 frontier AI systems from six developer teams. Despite recent advances in AI, no single model dominates all tasks, with omics-based age prediction being the hardest task regardless of scale. To test whether these gaps can be closed without frontier-scale resources, we fine-tuned a family of five multitask Longevity-LLMs on domain-specific aging data. The compact (0.6B-9B parameters) Longevity-LLMs matched or exceeded far larger frontier systems on LongevityBench, showing that general-purpose language models can be adapted to structured-omics tasks. We publicly release the benchmark, models, and Longevity Claw, an agentic research interface for aging researchers.

Aging↗

PubMed2026

AI-driven CRISPR screening: optimizing gene editing through automation and intelligent decision support.

BACKGROUND: CRISPR-based genetic screening has become a central methodology in functional genomics, enabling systematic interrogation of gene function, genetic interactions and context-dependent vulnerabilities at scale. However, the rapid expansion of screening modalities-including multi-condition designs, combinatorial perturbations, in vivo applications and single-cell readouts-has exposed fundamental limitations of heuristic-driven experimental design and post hoc statistical analysis. MAIN BODY: This Review synthesizes how artificial intelligence is reshaping CRISPR screening by introducing predictive, adaptive and system-level intelligence across the experimental lifecycle. We organize recent advances into two tightly coupled modules. First, machine learning and deep learning (ML/DL) methods optimize experimental design by learning context-dependent perturbation behavior, anticipating confounding effects and enabling iterative, information-efficient screening strategies. Second, large language model-agent (LLM-agent) systems complement these advances by externalizing scientific reasoning, integrating biological knowledge at scale and coordinating analysis and decision-making in human-in-the-loop workflows. CONCLUSIONS: Together, ML/DL and LLM-agent approaches reframe CRISPR screening from a static analytical pipeline into an intelligent experimental system, with important implications for robustness, scalability and biological discovery.

Artificial Intelligence↗

PubMed2026

scGPA: an LLM-assisted workflow for directional virtual gene perturbation analysis from single-cell transcriptomes.

BACKGROUND: Existing virtual perturbation methods can often infer directional changes by comparing predicted post-perturbation expression profiles with control cells. However, workflows that directly return direction-specific downstream candidate genes together with confidence scores, evidence support and interpretable summaries remain limited. We developed scGPA, an LLM-assisted workflow system for directional single-cell virtual gene perturbation analysis. METHODS: scGPA starts from raw single-cell RNA sequencing data and performs quality control, normalization, dimensionality reduction, clustering and cell-group selection. It then constructs cell-group-specific wild-type regulatory networks using repeated subsampling, principal component regression (PCR)/Ridge-based network inference and CP tensor denoising. Based on these networks, scGPA simulates dose-aware virtual knockdown of the target gene and applies signed perturbation propagation to estimate the magnitude and direction of downstream transcriptional responses. LLM assistance is used for marker-based cell-type annotation, evidence-guided candidate prioritization and user-facing biological summarization. RESULTS: We benchmarked scGPA across five public Perturb-seq datasets and compared its performance with GEARS, scGPT and a random baseline. The overall correct prediction rate of scGPA was 23.0%, exceeding those of GEARS (20.7%), scGPT (15.1%) and the random baseline (13.6%). These results indicate that scGPA achieved a higher correct prediction rate than the two comparator models and the random baseline. We subsequently evaluated scGPA using a public osteosarcoma single-cell dataset and performed qRT-PCR validation in 143B osteosarcoma cells. Among genes with significant experimental changes, scGPA achieved a directional concordance of 76.9%. When all tested downstream genes were counted, 37.0% were directionally correct, 51.9% showed no significant change and 11.1% changed in the opposite direction. CONCLUSIONS: scGPA provides a practical workflow system for predicting and prioritizing direction-specific downstream transcriptional responses after target-gene perturbation. By integrating single-cell regulatory network inference, signed virtual perturbation and LLM-assisted interpretation, scGPA supports target-gene function inference and downstream mechanistic investigation from single-cell transcriptomic data.

Single-Cell Gene Expression Analysis↗

PubMed2026

Biological Foundation Models for Complex Disease Research and Clinical Translation.

Complex diseases, including cancer, rare genetic disorders, neurodevelopmental and psychiatric conditions, and neurodegenerative diseases, arise from interactions among genetic variation, gene regulation, and cellular states that are difficult to capture using a single data type or biological scale. Biological foundation models address this challenge by treating nucleotides and genes as tokens and learning representations that can be transferred to downstream biomedical and clinical tasks. In this review, we examine two major model classes, genomic sequence foundation models and cell foundation models, and compare their tokenization strategies, model architectures, pretraining objectives, and adaptation methods. We summarize their emerging applications in regulatory variant interpretation, disease-associated cell-state analysis, drug-response prediction, and therapeutic target discovery across complex diseases. We distinguish applications supported by experimental or retrospective validation from those that remain primarily computational or conceptual. We further discuss key challenges to clinical translation, including multimodal data integration, model interpretability, benchmarking, patient-specific prediction, and privacy protection. We highlight future opportunities to integrate biological foundation models with emerging frameworks of medical digital twins, agentic AI, and federated learning. By linking model design to translational goals, this review provides a practical framework for evaluating biological foundation models and their readiness for complex disease research and clinical use.

biological foundation model↗

PubMed2026

LAMBDA: a prophage detection benchmark for genomic language models.

Transformer-based genomic sequence models represent an emerging frontier in computational biology. Yet, their embeddings have not yet shown the same level of predictive power as natural and protein language models, highlighting a gap between current implementations and theoretical promise. Existing benchmarks for DNA language models primarily focus on classifying regulatory elements in eukaryotic genomes, leaving open the fundamental question of whether these models learn sequence-level features across whole genomes. We introduce LAMBDA, a benchmark designed to rigorously evaluate genome language model embeddings through phage-bacteria sequence discrimination across four categories of increasing complexity: probing tasks, fine-tuning assessments, diagnostic tests, and genome-wide prophage detection. Our comprehensive analysis of current genomic language models provides insight into the importance of training data selection relative to model size, the need for domain-specific training, and the capabilities and limitations of genomic language models for detecting prophage sequences. This benchmark represents a challenging genomic annotation task in the bacterial domain and addresses a key computational problem with direct relevance to microbiology and medicine.

Prophages↗