搜索结果：token

The emergence of bacterial resistance to antibiotics poses a significant threat to human health; thus, there is an urgent need for new strategies in understanding the mechanisms of resistance and further fast prediction of it. Deep learning models offer promising solutions through analyzing genetic sequences in the prediction of bacterial resistance patterns. This study develops and validates a transformer-based deep learning model, DNABERT-2-117M, to predict gentamicin resistance in Klebsiella pneumoniae directly from whole-genome sequences. Our central methodological advance investigates the impact of the DNA tokenization strategy on predictive performance. We prospectively compared a dynamic tokenization approach against conventional fixed-length tokenization. Evaluated through rigorous fivefold cross-validation and on a hold-out test set, the model employing dynamic tokenization achieved superior performance, with a mean F1-score of 0.95 and an area under the curve of 0.97. Our findings establish that optimizing sub-sequence tokenization is crucial for model accuracy, and this dynamic tokenization approach significantly enhances model accuracy for antibiotic resistance prediction from genomic data. This genome-based predictive model represents a scalable and rapid alternative to traditional antibiotic susceptibility testing, offering the potential to accelerate clinical decision-making and improve patient outcomes in managing K. pneumoniae infections.IMPORTANCEThis study addresses a critical gap in diagnostic technologies for hypervirulent, antibiotic-resistant Klebsiella pneumoniae. We introduce a transformer-based deep learning framework that utilizes dynamic tokenization strategies to predict drug resistance directly from genomic sequences. The core significance of our work is the development of a robust genomic prediction model that serves as a foundational component for future diagnostic paradigms. The significance of this work lies in its potential to fundamentally alter clinical timelines. By decoupling resistance prediction from the requirement for phenotypic growth, our approach is a critical step toward next-generation workflows (e.g., clinical metagenomics) that could deliver a complete diagnostic and susceptibility report directly from a patient sample within hours. This represents a scalable, rapid diagnostic platform that promises to accelerate the administration of targeted treatment for high-risk K. pneumoniae infections, with a clear trajectory toward same-day, specimen-to-result diagnostics in the near future.

DNA replication is a biological process in which a single DNA molecule is duplicated, initiating from multiple genomic sites known as replication origins. Identifying replication origins and analyzing their underlying base sequence composition is crucial for understanding the mechanisms of DNA replication. Although there are various machine learning and deep learning approaches for origin prediction, many rely on labor intensive feature engineering or lack interpretability. We fine-tune two genome-based pretrained language models, DNABERT and DNABERT-2, to predict replication origins in budding yeast and unravel the DNA base composition behind them. The key contribution of this study is a systematic framework for analyzing genomic language models for replication origin prediction, combining controlled dataset design with model-specific explainability pipelines to examine how different tokenization strategies influence learned sequence features and whether such approaches can highlight biologically meaningful signals. We evaluate both models on the designed datasets to ensure robustness and support explainability. DNABERT demonstrates consistent performance, achieving an average accuracy of 0.72 for more challenging and 0.83 for the easier dataset. In comparison, DNABERT-2 achieved comparable scores of 0.72 and 0.81 on the same datasets. Our attention-based motif discovery pipeline enhances the interpretability of DNABERT, by identifying motifs from high-attention fragments that closely match known sequence patterns of replication origins. Perturbation-based explanation methods, including Shapley additive explanations, were applied to interpret DNABERT-2's learning mechanism. This analysis identified tokens with high attribution scores aligned with biologically relevant sequence composition. Our study demonstrates that both models identify replication origin sequences, albeit through different learning strategies. Tokenization appears to influence model learning and attention behavior in these models. The overlapping k-mer tokenization used in DNABERT yields more interpretable attention maps compared to the byte pair encoding tokenization employed in DNABERT-2. We show that despite sharing the same BERT-style architecture, DNABERT captures relevant short-range patterns and some sequence dependencies beyond just local context, as reflected in its attention maps. In contrast, DNABERT-2's alternative tokenization strategy biases its learning toward relevant short-range patterns by optimizing token weighting.

Test-time scaling has emerged as a promising method to enhance the reasoning capabilities of large language models (LLMs) and vision-language models (VLMs) during inference without additional training. While foundational studies established scaling paradigms in general domains, their applicability to the unique complexities of medical AI remains underexplored. This study aims to conduct a comprehensive investigation of test-time scaling in the medical domain. We evaluate the impact of scaling across different model sizes and task complexities. Furthermore, we seek to identify domain-specific bottlenecks and assess model robustness against user-driven perturbations, such as misleading clinical authority. This study evaluated a diverse set of general and medical-specific LLMs and VLMs. Experiments used five textual medical benchmarks comprising over 5500 questions and two multimodal benchmarks comprising 7000 samples. Performance was measured under three scaling conditions: increasing token budgets, iterative sequential scaling, and parallel scaling. Robustness was tested by embedding misleading hints with varying tones and levels of simulated clinical expertise into prompts. For nonreasoning LLMs, accuracy saturated quickly, with token usage often remaining under 500 tokens regardless of budget increases. Reasoning models demonstrated significant performance gains on complex tasks as token budgets increased. Notably, we identified distinct domain-specific behaviors. First, current VLMs showed a structural bottleneck in integrating visual clues and experienced limited benefit from token expansion. Second, medically fine-tuned LLMs excelled in clinical question answering but exhibited degraded scaling efficiency on calculation tasks compared to general-domain models. This reflects a disparity between qualitative clinical alignment and procedural logic. Third, while optimal scaling improved robustness, models exhibited a cognitive vulnerability by readily abandoning correct reasoning when confronted with misleading expert physician hints. Regarding scaling strategies, parallel scaling outperformed sequential scaling on easier tasks. Conversely, extended sequential scaling or increased budgets proved essential for complex problem-solving. Test-time scaling rules from general domains do not perfectly translate to medical AI. Longer reasoning is not universally beneficial. Concise reasoning with parallel scaling is optimal for simpler tasks. An extended chain of thought via sequential scaling or increased budgets is required for complex problems. Furthermore, safe clinical deployment requires addressing fundamental vision-language alignment, balancing clinical and procedural reasoning, and mitigating vulnerabilities to perceived clinical authority.

Most hallucination mitigation for large language models (LLMs) operates post-hoc, leaving safety-critical clinical deployment without real-time warning capability. We present a calibrated hidden-state probing pipeline that enables token-time hallucination detection under explicit false-positive-rate (FPR) constraints, making it suitable for streaming clinical deployment. We formulate token-time hallucination detection as a constrained sequential decision problem separating non-circular supervision construction (using ROUGE-L, exact-choice, and exploratory NLI verifiers), FPR-constrained operating-point selection, and downstream intervention evaluation into a modular, reusable protocol. A lightweight two-layer hidden-state probe classifies each generated token into one of three risk states (Safe, AtRisk, or Hallucinating), where the intermediate AtRisk state captures pre-error instability. We validate across four medical QA benchmarks (Endoscopy, PubMedQA, MedHallu, MedQA-USMLE) spanning approximately 44,500 generation trajectories and three backbone LLMs (Qwen3-8B, Llama-3.1-8B-Instruct, BioMistral-7B). On the public biomedical QA benchmarks the probe yields viable detection across all backbone-dataset combinations: EDR@5 up to 0.390 (PubMedQA) and 0.352 (MedHallu) at the strict cap γ≤0.10, rising to 0.587 with 29%-59% hallucination reduction at the monitoring cap γ≤0.20. The trigger fires roughly 10-35 tokens before error onset, whereas a post-hoc check has zero lead time by construction. A single-institution Endoscopy corpus serves as a case study (EDR@5 0.613, Llama-3.1), treated as illustrative because its small confirmed-correct denominator (26-52 per backbone) yields wide confidence intervals. Probe latency is ∼10ms per answer versus ∼10s for multi-sample baselines. Hidden-state probing with FPR-constrained calibration provides a practical, low-latency solution for real-time hallucination monitoring in clinical LLM deployments. The modular pipeline separating supervision construction, probe training, and trigger calibration is directly reusable with alternative detectors or verifiers.

Referring Video Object Segmentation (RVOS) aims to segment the target objects specified in human instructions. Previous approaches typically rely on explicit human instructions that contain target categories or salient appearance descriptions. These approaches tend to fail when the instructions require temporal video understanding and complex relational reasoning. In this work, we present RViSeg, a reasoning-centric video object segmentation model that leverages the reasoning capability of Multi-modal Large Language Models (MLLM) to handle complex queries. The primary challenge lies in enabling MLLM to perform efficient pixel-level video perception. To tackle this challenge, we introduce a novel Spatial Token Merge (STM) module that consolidates lengthy video tokens into compact region-level clusters, while preserving essential spatial details. This structured representation enables MLLM to infer user intention by interleaving spatial and temporal visual information. Furthermore, we propose a Query-based Target Retrieval (QTR) module that utilizes learnable tokens as the target identity for mask prediction. By propagating these instance-specific tokens both intra-clip and inter-clip, our RViSeg effectively encodes object motion, ensuring spatio-temporal consistency in segmentation results. To facilitate training and evaluation, we construct InstructVideo, a single- and multiple-object reasoning video segmentation benchmark. Comprehensive experiments demonstrate the effectiveness of the proposed components.

Fine-Grained Visual Classification (FGVC) remains a challenging task due to subtle inter-class differences and large intra-class variations. Existing approaches typically rely on feature-selection mechanisms or region-proposal strategies to localize discriminative regions for semantic analysis. However, these methods often fail to capture discriminative cues comprehensively while introducing substantial category-agnostic redundancy. To address these limitations, we propose $\text {H}^{3}$ Former, a novel token-to-region framework that leverages high-order semantic relations to aggregate local fine-grained representations with structured region-level modeling. Specifically, we propose the Semantic-Aware Aggregation Module (SAAM), which exploits multi-scale contextual cues to dynamically construct a weighted hypergraph among tokens. By applying hypergraph convolution, SAAM captures high-order semantic dependencies and progressively aggregates token features into compact region-level representations. Furthermore, we introduce the Hyperbolic Hierarchical Contrastive Loss (HHCL), which enforces hierarchical semantic constraints in a non-Euclidean embedding space. The HHCL enhances inter-class separability and intra-class consistency while preserving the intrinsic hierarchical relationships among fine-grained categories. Comprehensive experiments conducted on four standard FGVC benchmarks validate the superiority of our $\text {H}^{3}$ Former framework. Code is available at https://github.com/xiaozhangfangyang/H3Former.

Transformer-based models have significantly advanced image captioning through self-attention mechanisms and parallel computation. However, existing methods typically adopt teacher-forcing strategies during training by conditioning the decoder exclusively on ground-truth tokens, whereas at inference, captions are generated autoregressively based solely on previously predicted tokens. Such discrepancy between training and inference conditions leads to a progressive accumulation of prediction errors, resulting in captions that deviate significantly from the visual content. While scheduled sampling strategies mitigate this issue, directly integrating them disrupts Transformer parallelism and overlooks visual saliency differences. To tackle these limitations, we present VMFormer, a multi-stage decoding Transformer featuring a Visual-Aware Scheduled Sampling (VASS) module that bridges this gap through two key innovations: 1) A two-stage decoding scheme where an initial self-study stage generates candidate tokens, followed by a hybrid stage dynamically blending ground-truth references with predictions via a visual clues controlled gate. 2) A cognitive-inspired prioritization mechanism that retains visual keywords (nouns/verbs/attributes) in early training phases before transitioning to linguistic refinements, mirroring human captioning patterns. Crucially, the VASS module preserves the parallel computational strengths of Transformer architectures and is designed as a plug-and-play component, readily adaptable to Transformer-based captioning models. Experiments on challenging MSCOCO achieve 142.2 CIDEr. We extend VMFormer to video captioning and demonstrate consistent improvements on MSRVTT and MSVD datasets.

Despite the impressive generative capabilities of text-to-image (T2I) diffusion models, they remain vulnerable to implicit sexual prompts, where subtle cues disguised as benign terms or adversarial tokens unexpectedly generate the inappropriate content due to model biases or latent correlations in training data. Existing safety mechanisms face fundamental limitations: detection methods primarily identify explicit content and fail to capture implicit malicious intent, while mitigation approaches rely on static negative prompts inadequate for diverse implicit scenarios. To address these challenges, we propose UniNDM, a unified noise-driven framework that rethinks safety mechanisms through the lens of noise dynamics in diffusion processes. Our key insight is that early-stage predicted noise exhibits inherent separability between normal and sexually explicit content, which we theoretically demonstrates quadratically increasing semantic concentration with timestep. Leveraging this property, we develop a lightweight noise-based detector achieving superior accuracy with virtually no computational overhead. For mitigation, we introduce noise-enhanced adaptive negative guidance: dynamically generating context-specific negative prompts via large language models to handle diverse implicit content, while optimizing initial noise by suppressing attention concentration on explicit tokens to provide comprehensive protection. Besides the U-Net-based diffusion models, we further extend our framework to emerging Diffusion Transformer architectures through region-constrained semantic guidance tailored for their unified multimodal attention. Comprehensive experiments across U-Net models (SDv1.4, v1.5, v2.1, XL) and DiT models (SDv3) on both natural and adversarial datasets demonstrate substantial improvements over state-of-the-art methods, including SLD, UCE, Safree, etc. Our code is publicly available at https://github.com/Aries-iai/UniNDM.

Anxiety disorders are highly prevalent among adults with autism, with 20%-65% experiencing at least one diagnosable anxiety disorder. While mindfulness-based interventions have demonstrated efficacy for anxiety reduction, treatment response varies considerably across individuals. Machine learning approaches offer potential for identifying who is most likely to benefit from smartphone-based mindfulness interventions, enabling personalized treatment recommendations. This study aimed to develop and evaluate machine learning models to predict individual treatment response to a smartphone-based mindfulness intervention for adults with autism. We identified baseline characteristics that distinguish responders from nonresponders and explored few-shot learning with large language models (LLMs) as a complementary approach for low-data clinical prediction. We conducted a secondary analysis of a randomized controlled trial comparing a 6-week smartphone-based mindfulness intervention with a waitlist control group in adults with autism. Among 73 participants who completed the intervention, we defined responders as those achieving a &#x2265;7-point reduction in State-Trait Anxiety Inventory state anxiety scores. Baseline predictors included demographic variables; autism trait measures; and self-report questionnaires assessing anxiety symptoms, perceived stress, affect, and mindfulness. To determine which machine learning model was most predictive of response, we trained 6 different models (logistic regression, random forest, extreme gradient boosting [XGBoost], tabular data network [TabNet], TabICL, and Tabular Prior-Data Fitted Network [TabPFN]) using nested 10-fold cross-validation with inner 5-fold cross-validation for hyperparameter tuning and evaluated GPT-4o few-shot learning with tokenized features at 20 to 70 shots. Random forest achieved the highest predictive performance for state anxiety response (area under the curve [AUC] 0.79, 95% CI 0.66-0.91), followed by TabPFN (AUC 0.78, 95% CI 0.64-0.94) and logistic regression (AUC 0.77, 95% CI 0.73-0.81). Higher baseline state anxiety (standardized &#x3b2; coefficient=1.20, P<.001) predicted better treatment response, while higher Autism Spectrum Quotient at baseline (standardized &#x3b2; coefficient=-0.17, P=.001), older age (standardized &#x3b2; coefficient=-0.18, P=.02), and lower childhood pretend play scores (standardized &#x3b2; coefficient=-0.93, P=.007) were associated with poorer response. Few-shot learning with 7-feature tokenization achieved an accuracy of 0.867 at 70 shots, compared to an accuracy of 0.733 for random forest. Prediction of trait anxiety changes was substantially weaker (AUCs 0.46-0.68), likely reflecting the inherent stability of this personality dimension. Machine learning models successfully identified baseline characteristics predicting state anxiety response to a smartphone-based mindfulness intervention in adults with autism. Few-shot learning with LLMs demonstrated superior performance to traditional machine learning when provided with compact, high-signal feature representations, offering a promising approach for clinical prediction in small-sample settings. These findings demonstrate the feasibility of precision psychiatry in digital mental health interventions for adults with autism. As online mental health interventions become ubiquitous, patients and clinicians can know whether a particular intervention is more or less likely to benefit an individual patient.

Existing image-text retrieval methods often compute cross-modal similarity using global single-vector representations. Although efficient for coarse semantic alignment, such compressed representations are limited when textual queries involve fine-grained semantics, including objects, attributes, relations, and their compositional structures. This paper focuses on fine-grained text-to-image retrieval and proposes Q-GrAM, a retrieval-oriented adaptation of the BLIP-2 Q-Former. Instead of treating Q-Former queries as a homogeneous set, Q-GrAM partitions a fixed query budget into semantically differentiated groups. A text-guided router assigns token-level semantic demands to query groups, while query conditional initialization modulates each group according to group-level textual summaries. The resulting grouped visual query features are matched with text tokens through a group-aware late interaction scorer, and auxiliary routing balance and inter-group diversity regularization are introduced to stabilize semantic specialization. Experiments on MS-COCO 5K, Flickr30K, and Flickr30K-CFQ show that Q-GrAM achieves strong text-to-image retrieval performance against both global embedding baselines and representative fine-grained image-text matching methods, while maintaining competitive bidirectional retrieval performance. These results demonstrate the effectiveness of structured, text-conditioned Q-Former query specialization for fine-grained text-driven image search.

The power-law relationship between large language model (LLM) performance and the scale of pre-training corpora has driven remarkable advances, while raising serious ethical and legal concerns. Pre-training corpora often contain sensitive personal information, copyrighted content, or benchmark test data, and the opacity of pre-training corpora further exacerbates these concerns. Consequently, detecting pre-training data has become a significant research challenge. Most existing detection methods rely on intermediate outputs of LLMs (activations, token probabilities, or model loss), which are inaccessible in commercial LLMs, where only final outputs are available. Only a few early-stage attempts rely solely on final outputs by measuring lexical or semantic similarity between reproduced and original texts at the sentence or token level. Yet, these attempts overlook the dynamic nature of reproducibility: reproducibility fluctuates differently across reproduction positions, contextual spans, and expressions, between seen and unseen texts. This oversight narrows the gap of reproduction similarity, ultimately degrading detection performance. To address this limitation, we propose RADS-PDD (reproducibility-aware dynamic similarity-based pre-training data detection), which shifts black-box pre-training data detection from static similarity comparison to reproducibility-aware dynamic similarity modeling. RADS-PDD incorporates three reproducibility-aware mechanisms, positional gain weight, continual gain weight, and triplet occurrence probability, to quantify varying reproducibility by dynamic similarity, amplifying the reproduction similarity gap between seen and unseen texts. Extensive experiments across two representative datasets, multiple data domains, and diverse LLMs demonstrate that RADS-PDD consistently outperforms detection methods that rely solely on final outputs and achieves performance comparable to detection methods that require intermediate outputs.

Multi-modal object Re-Identification (ReID) aims to retrieve specific objects by integrating complementary information from multiple modalities. However, existing multi-modal ReID methods do not effectively address background interference suppression or achieve tri-modal alignment, instead focusing on pairwise feature fusion. Moreover, many current aggregation approaches suffer from high computational complexity. To address these limitations, we propose PRISM, a novel multi-modal ReID framework built upon Prompt-S6 (PS6) and semantic-aware knowledge guidance. PS6 maintains the linear complexity and strong sequence modeling capability of Mamba while enabling efficient cross-modal interaction. Leveraging these advantages, we design two key components: Semantic-Driven Token Pruning (SDTP) and Progressive Fusion Network (PFN). Parsing semantic priors from the segmentation foundation models, the SDTP then leverages these priors and applies dynamic token pruning to suppress background noise and refine feature representations. The PFN progressively aggregates multi-modal features to achieve tri-modal alignment and fully exploit modality complementarity. With the proposed modules, PRISM generates more robust multi-modal representations under complex scenarios. Extensive experiments on four multi-modal object ReID benchmarks demonstrate the effectiveness and efficiency of our approach. The source code is available at https://github.com/zw-absin/PRISM.

Motion quantization codebooks have been widely adopted to facilitate co-speech motion generation. However, the conventional quantization-based generation paradigm-which relies on probabilistic token sampling from limited discrete code-books-suffers from two major limitations: crude, unreasonable motion representations and fixed, homogenized motion token sequences. To overcome these issues, we propose a novel explicit generation paradigm based on generative continuous quantization. Specifically, we first introduce a continuous quantization method to derive a set of generative motion units. This approach enables smoother and more accurate representation of human motion compared to classical methods. Building on these generative units, we further propose a compositional weight generation paradigm that replaces probabilistic sampling with deterministic, explicit motion synthesis. Moreover, as generalization capability is crucial for real-world deployment, we design a fully audio-aware encoder to extract style features that are decoupled from content. These features are integrated into the motion decoder via Adaptive Instance Normalization to enhance cross-speaker facial style generalization. Our method achieves state-of-the-art performance on two public datasets. Notably, owing to its concise and efficient architecture, our model attains an inference speed exceeding 4000 fps on the SHOW dataset, demonstrating strong potential for practical real-time applications.

This study combines modern source-separation techniques with interpretable machine learning methods to investigate how binary perceptual annotations of diphthongal and monophthongal /a&#x26a;/ relate to measurable acoustic variation in sung vocal performance. Using a corpus of studio-recorded music processed with vocal-instrumental source separation, we extract F1 and F2 trajectories for 1004 /a&#x26a;/ tokens, each perceptually categorized as either monophthongal or diphthongal. These trajectories were modeled using two approaches: (i) a gradient boosted decision tree trained on engineered acoustic features and (ii) a multilayer perceptron (MLP) trained directly on raw formant trajectories. Both models achieved high accuracy and Area Under the Receiver Operating Characteristic Curve, indicating that perceptual labels can reliably be predicted from both raw and engineered acoustic input. Moreover, a Shapley Additive Explanations-based feature importance analysis showed that features such as &#x394;F1/&#x394;F2, cubic spline coefficients, and trajectory derivatives captured systematic differences between perceived monophthongal and diphthongal tokens, highlighting the value of dynamic representations over static distance-based measures. The results indicate that perceptual annotations of diphthongal and monophthongal /a&#x26a;/ correspond with quantifiable acoustic information and demonstrate how explainable machine learning can help map gradient vowel dynamics onto binary perceptual categories. The study further emphasizes how recent advances in source separation make sung performance a valuable new domain for phonetic research.

Demographic variables such as age, education, sex, and race/ethnicity influence performance on neuropsychological tests. We present a proof-of-concept comparison of two approaches to incorporating race/ethnicity in regression-based norms for the Meyers Neuropsychological Battery (MNB). Linear regression models were developed on a reference dataset (N&#x2009;=&#x2009;1,840) to predict T-scores from demographic predictors and raw scores, using either the MNB's existing single ordinal race/ethnicity variable or four indicator-coded variables. Models were applied to an independent sample (N&#x2009;=&#x2009;10,778) to generate predicted scores and correction factors (actual minus predicted T-score). After Benjamini-Hochberg correction across 34 measures, only 3 (Trails B, Token Test, and Boston Naming) showed significant improvement with indicator coding. Delta R-squared analyses indicated that race/ethnicity indicators contributed 0.29% to 0.45% of unique variance. Correction factors for certain test-by-group combinations approached one standard deviation, but bootstrap confidence intervals revealed substantial instability attributable to small minority cell sizes in the development sample (e.g., N&#x2009;=&#x2009;22 African American). These results suggest that race/ethnicity is a weak overall predictor once other demographics are controlled, though systematic prediction biases may exist for specific measures. The indicator-coded framework can in principle accommodate mixed racial/ethnic backgrounds, though this was not tested here.

Accurate prediction of molecular bioactivity is a fundamental goal in rational drug design but remains challenging due to data scarcity and label imbalance. To address these limitations, we propose a unified Bayesian framework that integrates classification-to-regression knowledge fusion, uncertainty quantification, and active learning for data-efficient molecular property prediction. Specifically, we develop the Bayesian Class-Attentive Transformer Network (BCATNet). This model learns activity patterns from abundant classification data and incorporates the predicted probabilities as informative priors to guide the subsequent Bayesian regression task. Structurally, BCATNet employs a cross-token attention mechanism to model nonlinear interactions between class-derived semantics and molecular structural features. Comparative experiments against conventional machine learning models, graph neural networks, pretrained molecular models, and classification-guided baselines further demonstrated that explicit classification-to-regression knowledge fusion can provide a competitive and data-efficient alternative to generic molecular pretraining. Under reduced regression supervision, BCATNet maintained lower prediction errors and stronger robustness than competing models, supporting its utility in label-scarce settings. Beyond accuracy, the Bayesian formulation generated uncertainty estimates that were informative for reliability assessment: high-uncertainty predictions showed larger regression errors, and uncertainty-based risk stratification separated low-, medium-, and high-risk molecular predictions. Finally, BCATNet uncertainty served as an effective acquisition signal in active learning, with uncertainty-driven strategies achieving the best final performance in most benchmark tasks. Overall, BCATNet establishes a generalizable paradigm for uncertainty-aware molecular modeling by bridging classification and regression tasks within a Bayesian framework, offering a principled route toward reliable, interpretable, and resource-efficient drug discovery.

Genomic language models (gLMs) are rapidly becoming important tools for learning biological information directly from sequence data. By adapting concepts from natural language processing, these models aim to capture contextual dependencies, regulatory grammar, evolutionary constraint, and sequence-level functional patterns that may be difficult to detect using alignment-based, motif-based, or conventional supervised methods alone. This systematic review evaluates recent model-development studies of genomic, RNA, nucleotide, codon-level, and regulatory DNA language models, with emphasis on model architecture, tokenization, training objective, biological task, benchmarking strategy, and reported limitations. A structured search of PubMed, Scopus, and Web of Science identified 469 records. After duplicate removal, screening, and full-text eligibility assessment, 58 studies met the strict inclusion criteria for primary model development or substantial model adaptation. The included studies covered diverse applications, including regulatory sequence prediction, variant-effect modeling, genome annotation, microbial and viral genome analysis, RNA splicing and regulation, codon optimization, mRNA design, and generative design of regulatory or RNA sequences. Across the included studies, stronger evidence for gLM utility was generally associated with biologically informed or task-aligned model design, including evolutionary alignments, motif-aware objectives, long-context architectures, RNA structural priors, population-aware representations, and domain-specific pretraining. However, the evidence was heterogeneous and did not support a general claim of superiority over established bioinformatics tools or specialized supervised models. In several regulatory genomics tasks, specialized supervised models, k-mer-based approaches, or conventional deep-learning baselines remained competitive or superior to pretrained language-model representations. Generative models showed growing promise for RNA, codon, viral genome, and cis-regulatory element design, although many were evaluated mainly in silico. Overall, the field is advancing quickly, but broader impact will require standardized benchmarks, clearer reporting, stronger external validation, improved interpretability, and experimental confirmation of predicted or generated biological functions.

Large language models (LLMs) are now embedded in scientific, educational, and governance workflows, with debates centering on their capabilities, mechanisms, and impacts. Yet these debates remain structured by persistent folk theories-intuitive, informal explanatory models that guide attitudes and actions. Deflationary slogans ("just autocomplete," "stochastic parrots," and "average of the internet") and anthropomorphic framings ("emergent agents" and "proto-minds") each capture genuine features of current systems but mistake those features for the whole. This Perspective proposes a minimal working model of LLM-based systems centered on four distinctions: between pretraining and deployed systems; between the learned distribution and particular samples; among parametric, contextual, and external memory; and between task competence and agency. The model is used to diagnose six misconceptions about LLMs: next-token prediction, regression to the mean, training-data regurgitation, model memory, alignment, and understanding. For each, the analysis identifies what the misconception gets right, which distinctions it conflates, and what follows for capability evaluation, system design, and governance. Applied to publisher AI policies as governance case studies, the framework shows both how policy language can conflate these distinctions and how such errors can be corrected. The model thereby avoids the parrot-mind binary by treating LLMs as simulators of discourse and task performance, offering a diagnostic toolkit for locating and correcting the errors these folk theories perpetuate.