Zongliang Yue

Ribosome profiling enables genome-wide measurement of translation at nucleotide resolution and provides a dynamic view of cellular protein synthesis under diverse biological conditions. Existing computational approaches primarily operate on codon-level representations, potentially losing fine-grained translational signals critical for modeling context-dependent cellular responses. Such predictive translational modeling is increasingly important for emerging biological digital twins, where accurate simulation of molecular-state dynamics is required to characterize cellular adaptation, perturbation response, and phenotype progression. We present RiboRep, a replicate-aware cross-modal transformer for codon-resolved ribosome density prediction. RiboRep jointly models nucleotide-resolution RNA sequences and reference ribosome occupancy signals using dual-stream convolutional encoders, RoPE-based self-attention, asymmetric cross-attention, and replicate-aware conditioning tokens. By explicitly modeling replicate-specific variation and integrating sequence context with experimentally observed translational activity, RiboRep provides a framework for reconstructing and simulating translational states across biological conditions. Across bacterial, yeast, and plant ribosome profiling datasets, RiboRep achieves competitive or improved performance compared with existing baselines, with particularly strong gains on replicate-rich plant datasets. Ablation studies further demonstrate the importance of local codon-aware feature extraction, replicate-aware conditioning, and gated readout. Beyond predictive performance, the proposed framework establishes a foundation for translation-aware molecular digital twins capable of modeling ribosome occupancy landscapes, perturbation-induced translational responses, and condition-specific regulatory programs at codon resolution. ¹

SUMMARY: Gene Ontology encodes genes as a hierarchy, yet every enrichment visualization flattens it into a ranked list, discarding the ability to view the same process at different levels of abstraction. We present MondrianMap, a free interactive web application ( https://mondrianmap.smartdrugdiscovery.org/ ) that organizes enrichment results into 13 semantically principled layers derived from the GOALS framework and renders them as color-encoded rectangular maps where block area reflects significance, color encodes effect direction, and spatial proximity preserves semantic relations, all within an interactive interface. Three case studies across the NIH Common Fund Data Ecosystem demonstrate that visualization facilitates recognition of patterns that are difficult to discern in flat outputs: (1) LINCS CRISPR perturbations reveal that TP53 and KRAS knockouts produce opposite color maps at a single semantic layer, the same immune recruitment processes suppressed by TP53 loss are activated by KRAS disruption; (2) GTEx aging signatures expose the inflammaging paradox as an immediate visual phenomenon, identical antimicrobial defense programs appear uniformly upregulated in aging blood yet uniformly downregulated in aging liver at matched semantic resolution; and (3) MoTrPAC exercise data capture temporal dynamics as color transitions where brown adipose tissue undergoes a threshold switch from two enriched terms to thirty-four at a single molecular layer, while cardiac tissue reverses from uniformly activated to uniformly suppressed glycolytic metabolism as adaptation progresses. MondrianMap facilitates hierarchical visual reasoning, complementing statistical enrichment reporting for biological discovery. HIGHLIGHTS: MondrianMap provides multi-resolution enrichment visualization for Gene Ontology Layer-specific views reveal directional oppositions difficult to discern in flat term lists Navigating layers organizes one enrichment into a multi-scale biological narrative Demonstrated on cancer, aging, and exercise data across three CFDE programs IN BRIEF: MondrianMap is a web application that transforms gene set enrichment results into layered visualizations encoding regulation, significance, and semantic hierarchy. Across cancer, aging, and exercise datasets, viewing enrichment at defined semantic layers exposes directional inversions, temporal switches, and multi-scale biological narratives that are difficult to extract from conventional flat enrichment outputs. THE BIGGER PICTURE: When researchers measure gene expression changes in disease, aging, or drug response, they rely on enrichment analysis to translate thousands of molecular measurements into interpretable biological themes. The standard output is a ranked list of processes sorted by statistical significance. This format served the field well when studies examined a single condition; however, modern genomics routinely compares dozens of tissues, timepoints, and perturbations, each generating hundreds of enriched terms. The critical limitation is not statistical power; however, interpretive structure: a flat list cannot show whether two conditions activate the same biological process in opposite directions, whether a process visible at one level of abstraction disappears or transforms at another, or how a tissue’s functional response evolves across time. These are precisely the questions that define contemporary systems biology. Does a tumor suppressor gene silence the same immune program that an oncogene activates? Does aging drive the same defense pathway upward in the blood and downward in the liver? Does an exercise response flip from activation to suppression as the tissue adapts? Answering these questions requires a visualization framework that preserves hierarchy, encodes direction, and enables comparison at matched levels of biological resolution. MondrianMap provides this framework by organizing Gene Ontology terms into quantitatively defined semantic layers and rendering enrichment as color-encoded rectangular maps navigable from molecular mechanism to system-level theme. The result is a tool that extends enrichment analysis from a reporting step into an interactive framework for biological reasoning, hypothesis generation, and cross-dataset discovery.

Pathway analysis is a critical step for translating gene-level omics results into biological mechanisms, yet existing workflows often leave researchers with long lists of statistically significant pathways that are difficult to interpret, validate, and connect to experimental context. We developed PAG-Agent, a biologist-oriented virtual research assistant that integrates pathway-level statistical analysis, context-aware biological interpretation, literature-supported reasoning, and scientific writing support within a unified workflow. PAG-Agent supports bulk and single-cell transcriptomic data and enables users to perform data preprocessing, differential expression analysis, pathway analysis, pathway-level consensus analysis, and pathway-level meta-analysis through click-based and chat-based interactions. Unlike conventional pathway-analysis tools that analyze gene sets largely in isolation, PAG-Agent incorporates experimental conditions and research objectives to prioritize biologically relevant pathways and generate interpretable hypotheses. The system also provides gene and pathway annotation, citation retrieval, visualization, and writing refinement functions. In Alzheimer's disease case studies using three transcriptomic datasets, PAG-Agent consistently identified neurodegeneration-related pathways across multiple analysis methods and datasets. In citation-retrieval benchmarking, PAG-Agent outperformed six competing LLMs across five common literature-support scenarios, demonstrating improved ability to provide contextually relevant and valid references. Overall, PAG-Agent lowers technical barriers for pathway-level analysis and helps researchers move from transcriptomic data to biologically grounded interpretation, hypothesis generation, and scientific communication.

BACKGROUND The biomedical literature is expanding at an unprecedented rate, with over 4,000 new articles indexed on PubMed each day. Clinicians and researchers frequently lack the time to review this volume before making decisions. Retrieval-Augmented Generation (RAG) systems attempt to bridge this gap by grounding language model responses in relevant documents, but standard implementations rank all retrieved passages solely by semantic similarity, treating a case report and a meta-analysis as equally authoritative. OBJECTIVE This study aimed to develop and pilot-evaluate an evidence- and temporally aware retrieval-augmented generation framework that integrates evidence quality and publication recency into retrieval scoring. Using Alzheimer’s disease literature as a test case, we assessed whether incorporating these signals improved biomedical question-answering quality relative to conventional cosine-similarity RAG and a full-context baseline. METHODS We developed ET-RAG (Evidence-Temporal Retrieval-Augmented Generation), a retrieval framework that ranks each retrieved text chunk using a weighted score integrating cosine similarity (50%), evidence quality based on the GRADE hierarchy (30%), and temporal recency (20%). We evaluated ET-RAG against two baselines: a full-context agent powered by Gemini 2.0 Flash and a standard cosine-similarity RAG agent using GPT-4o-mini. To assess performance, we constructed a benchmark of 40 questions derived from 10 peer-reviewed Alzheimer’s disease papers published between 2021 and 2025, including 10 single-choice, 10 multiple-choice, 10 short-answer, and 10 long-answer questions. Performance was evaluated using correctness for choice-based questions and completeness, accuracy, and relevance for open-ended questions, with scoring conducted using an LLM-as-a-judge framework. RESULTS ET-RAG achieved the highest scores across all four question categories: single choice (0.90), multiple choice (0.74), short answer (0.92), and long answer (0.89), with a combined average of 0.86. Cosine RAG scored 80%, 0.48, 0.82, and 0.69, respectively (average 0.70), while the full context agent scored 0.60, 0.59, 0.71, and 0.53 (average 0.61). The full context agent, despite having access to the entire corpus through Gemini’s large context window, struggled with consistent answer extraction and was prone to rate limiting under heavy query loads. A control question on forestry was correctly rejected by all three agents, suggesting no hallucination on this control item. CONCLUSIONS In the Alzheimer’s disease benchmark, incorporating evidence quality and recency into RAG retrieval improved answer quality relative to pure cosine similarity retrieval and full-corpus prompting. The evidence-temporal scoring function is lightweight to implement and adds minimal computational overhead to existing vector search pipelines, but broader validation across domains, evidence levels, and stronger retrieval baselines are required before claims of generalizable biomedical reliability can be made.

Background: The biomedical literature is expanding at an unprecedented rate, with over 4,000 new articles indexed on PubMed each day. Clinicians and researchers frequently lack the time to review this volume before making decisions. Retrieval-Augmented Generation (RAG) systems attempt to bridge this gap by grounding language model responses in relevant documents, but standard implementations rank all retrieved passages solely by semantic similarity, treating a case report and a meta-analysis as equally authoritative. Objective: We aimed to develop and pilot-evaluate a RAG variant that incorporates evidence quality and publication recency into the retrieval scoring function, and to determine whether these signals improve answer quality on biomedical questions compared with standard cosine similarity RAG and a full-context baseline. Methods: We developed ET-RAG (Evidence-Temporal RAG), which scores each retrieved chunk using a weighted combination of cosine similarity (50%), evidence quality based on the GRADE hierarchy (30%), and temporal recency (20%). We evaluated ET-RAG alongside two baselines: a full context agent powered by Gemini 2.0 Flash and a standard cosine RAG agent using GPT-4o-mini. All agents were tested on 40 benchmark questions (10 single-choice, 10 multiple-choice, 10 short answer, and 10 long answer) drawn from 10 peer-reviewed Alzheimer's disease papers published between 2021 and 2025. Results: ET-RAG achieved the highest scores across all four question categories: single choice (0.90), multiple choice (0.74), short answer (0.92), and long answer (0.89), with a combined average of 0.86. Cosine RAG scored 80%, 0.48, 0.82, and 0.69, respectively (average 0.70), while the full context agent scored 0.60, 0.59, 0.71, and 0.53 (average 0.61). The full context agent, despite having access to the entire corpus through Gemini's large context window, struggled with consistent answer extraction and was prone to rate limiting under heavy query loads. A control question on forestry was correctly rejected by all three agents, suggesting no hallucination on this control item. Conclusions: In this pilot Alzheimer's disease benchmark, incorporating evidence quality and recency into RAG retrieval improved answer quality relative to pure cosine similarity retrieval and full-corpus prompting. The evidence-temporal scoring function is lightweight to implement and adds minimal computational overhead to existing vector search pipelines, but broader validation across domains, evidence levels, and stronger retrieval baselines are required before claims of generalizable biomedical reliability can be made.

1. Data is an important asset to improve healthcare through its utility in clinical studies and public health efforts. There is a need to access vast amounts of high-quality data to gather evidence...

The integrative analysis of gene sets, networks, and pathways is pivotal for deciphering omics data in translational biomedical research. To significantly increase gene coverage and enhance the utility of gene sets from diverse sources, we introduced pathways, annotated gene lists, and gene signatures (PAGs) enriched with metadata to represent biological functions. Furthermore, we established PAG–PAG networks by leveraging gene member similarity and gene regulations. However, in practice, high similarity in descriptions and gene membership often produces redundant, lengthy PAG lists, leading to gene set enrichment results that are difficult to interpret. We present Graph-Oriented Learning with Domain-Embedding Network (GOLDEN) fusion, an integrative framework that jointly leverages (i) connection-based embeddings derived from PAG–PAG relationships and (ii) semantic-based embeddings learned from PAG descriptions with a large language model (LLM). The two representations are combined via early fusion with a tunable weighting to produce a unified embedding on which clustering identifies concise, higher-level super-PAG. To assess when clustering is appropriate, we introduce a Connection Disparity Index, trained on synthetic stochastic block models, as a proxy for network “clusterability.” We further optimize the number of clusters with consensus clustering. On Gene Ontology Annotation biological process benchmarks, GOLDEN fusion recovers is-a structure more accurately than either connection-only or semantic-only baselines, demonstrating consistent gains in Adjusted Rand Index and Normalized Mutual Information. Finally, we generate summaries for each super-PAG by synthesizing its member PAG descriptions using a comparative analysis of LLMs. GOLDEN fusion provides an integrated framework for interpreting omics results.

Background: A major technical challenge in single-cell transcriptomics is the absence of an integrative analytic pipeline that can simultaneously leverage gene regulatory network (GRN) architecture, AI-assisted gene panel discovery, and functional relevance analyses to generate coherent biological insights. Existing approaches often treat these components independently, focusing on clusters, marker genes, or predictive features without integrating them into a mechanistically grounded framework. Consequently, comprehensive screening that links regulatory association, gene signature screening, and functional interpretation within single-cell datasets remains limited, underscoring the need for an integrated strategy. Methods: We developed an integrative bioinformatics pipeline based on Gene regulatory network–AI–Functional Analysis (GAFA), combining latent-space integration, unsupervised clustering, diffusion pseudotime analysis, lineage-resolved generalized additive modeling, GRN inference, and machine learning-based gene panel discovery. This framework enables systematic mapping of cell-state structure, reconstruction of differentiation and effector trajectories, and identification of transcriptional and regulatory features strongly associated with clinical outcomes. As a case study, we applied the pipeline to NK cell transcriptomes from six CML patients (two early relapse, two late relapse, two durable treatment-free remission—TFR; 15 samples) collected at TKI discontinuation and 6–12 months after therapy cessation. Results: We reanalyzed publicly available scRNA-seq data from a previously published CML cohort to evaluate NK-cell transcriptional programs associated with treatment-free remission and relapse. We resolved six transcriptionally distinct NK cell states spanning CD56^bright-like cytokine-responsive, early activated, terminally mature, cytotoxic, lymphoid trafficking, and HLA-DR⁺ immunoregulatory populations, each exhibiting outcome-specific compositional differences. Pseudotime analysis revealed two major NK cell lineages—a maturation trajectory and a cytotoxic effector trajectory. TFR samples displayed balanced occupancy of both lineages, whereas early relapse samples showed marked depletion of the maturation branch and preferential accumulation in cytotoxic end states. AI-guided feature selection and random forest modeling identified an 18-gene panel that distinguished NK cells from TFR and relapse samples in an exploratory manner. Among them, CST7, FCER1G, GNLY, GZMA, and HLA-C were conventional NK-associated genes, whereas ACTB, CYBA, IFITM2, IFITM3, LYZ, MALAT1, MT2A, MYOM2, NFKBIA, PIM1, S100A8, S100B, and TSC22D3 were novel. The GRN inference further uncovered outcome-specific regulatory modules, with RUNX3, EOMES, ELK4, and REL regulons enriched in TFR, whereas FOSL2 and MAF regulons were enriched in relapse, and their downstream targets linked to IFN-γ signaling, metabolic reprogramming, and immunoregulatory feedback circuits. Conclusions: This AI-enabled single-cell analysis demonstrates how NK cell state composition, differentiation trajectories, and regulatory network rewiring collectively shape TFR versus relapse following TKI discontinuation in CML. The integrative pipeline provides a modular framework that could be extended to additional datasets for data-driven biomarker discovery and mechanistic stratification, and highlights candidate transcriptional regulators and NK cell programs that may be leveraged to improve remission durability, pending validation in larger patient cohorts.