Sui Huang

Diatoms are the most successful and diverse phytoplankton group in the world's oceans. Here, we surveyed, at single-cell resolution, dynamic changes in the transcriptional landscape of a clonal population of Thalassiosira pseudonana as it transitioned through the light/dark cycle and fluctuations in nitrogen (N) availability conditions that simulated the end of a seasonal bloom. By profiling >7,000 single-cell transcriptomes, we have discovered extensive heterogeneity in the transcriptional states of co-existing T. pseudonana sub-populations within a clonal population. Our findings show how functionally distinct sub-populations drive adaptation of T. pseudonana to the light/dark cycle, while redistributing C and N during N-depletion. The diatom culture adapts to N-starvation through an increase in the relative size of a pre-existing sub-population expressing N-uptake and utilization genes, which enables its rapid recovery upon N-influx. We discuss how dynamic changes in the relative composition of functionally diverse sub-populations further demonstrate how anticipatory and bet-hedging strategies likely contribute to the ecological success of diatoms in the world’s oceans, by enabling their rapid adaptation to routine, as well as stressful, and unexpected environmental fluctuations.

The growing availability of biomedical data offers vast potential to improve human health, but the complexity and lack of integration of these datasets often limit their utility. To address this, the Biomedical Data Translator Consortium has developed an open‐source knowledge graph–based system—Translator—designed to integrate, harmonize, and make inferences over diverse biomedical data sources. We announce here Translator's initial public release and provide an overview of its architecture, standards, user interface, and core features. Translator employs a scalable, federated, knowledge graph framework for the integration of clinical, genomic, pharmacological, and other biomedical knowledge sources, enabling query retrieval, inference, and hypothesis generation. Translator's user interface is designed to support the exploration of knowledge relationships and the generation of insights, without requiring deep technical expertise and gradually revealing more detailed evidence, provenance, and confidence information, as needed by a given user. To demonstrate Translator's application and impact, we highlight features of the user interface in the context of three real‐world use cases: suggesting potential therapeutics for patients with rare disease; explaining the mechanism of action of a pipeline drug; and screening and validating drug candidates in a model organism. We discuss strengths and limitations of reasoning within a largely federated system and the need for rich concept modeling and deep provenance tracking. Finally, we outline future directions for enhancing Translator's functionality and expanding its data sources. Translator represents a significant step forward in making complex biomedical knowledge more accessible and actionable, aiming to accelerate translational research and improve patient care.

Genome sequencing of cancer and normal tissues, alongside single-cell transcriptomics, continues to produce findings that challenge the idea that cancer is a ‘genetic disease’, as posited by the somatic mutation theory (SMT). In this prevailing paradigm, tumorigenesis is caused by cancer-driving somatic mutations and clonal expansion. However, results from tumor sequencing, motivated by the genetic paradigm itself, create apparent ‘paradoxes’ that are not conducive to a pure SMT. But beyond genetic causation, the new results lend credence to old ideas from organismal biology. To resolve inconsistencies between the genetic paradigm of cancer and biological reality, we must complement deep sequencing with deep thinking: embrace formal theory and historicity of biological entities, and (re)consider non-genetic plasticity of cells and tissues. In this Essay, we discuss the concepts of cell state dynamics and tissue fields that emerge from the collective action of genes and of cells in their morphogenetic context, respectively, and how they help explain inconsistencies in the data in the context of SMT.

Large Language Models (LLMs) are being adopted at an unprecedented rate, yet still face challenges in knowledge-intensive domains like biomedicine. Solutions such as pre-training and domain-specific fine-tuning add substantial computational overhead, requiring further domain-expertise. Here, we introduce a token-optimized and robust Knowledge Graph-based Retrieval Augmented Generation (KG-RAG) framework by leveraging a massive biomedical KG (SPOKE) with LLMs such as Llama-2-13b, GPT-3.5-Turbo and GPT-4, to generate meaningful biomedical text rooted in established knowledge. Compared to the existing RAG technique for Knowledge Graphs, the proposed method utilizes minimal graph schema for context extraction and uses embedding methods for context pruning. This optimization in context extraction results in more than 50% reduction in token consumption without compromising the accuracy, making a cost-effective and robust RAG implementation on proprietary LLMs. KG-RAG consistently enhanced the performance of LLMs across diverse biomedical prompts by generating responses rooted in established knowledge, accompanied by accurate provenance and statistical evidence (if available) to substantiate the claims. Further benchmarking on human curated datasets, such as biomedical true/false and multiple-choice questions (MCQ), showed a remarkable 71% boost in the performance of the Llama-2 model on the challenging MCQ dataset, demonstrating the framework’s capacity to empower open-source models with fewer parameters for domain-specific questions. Furthermore, KG-RAG enhanced the performance of proprietary GPT models, such as GPT-3.5 and GPT-4. In summary, the proposed framework combines explicit and implicit knowledge of KG and LLM in a token optimized fashion, thus enhancing the adaptability of general-purpose LLMs to tackle domain-specific questions in a cost-effective fashion. SPOKE KG can be accessed at https://spoke.rbvi.ucsf.edu/neighborhood.html. It can also be accessed using REST-API (https://spoke.rbvi.ucsf.edu/swagger/). KG-RAG code is made available at https://github.com/BaranziniLab/KG_RAG. Biomedical benchmark datasets used in this study are made available to the research community in the same GitHub repository. Supplementary data are available at Bioinformatics online.

Poor prognosis and drug resistance in glioblastoma (GBM) can result from cellular heterogeneity and treatment-induced shifts in phenotypic states of tumor cells, including dedifferentiation into glioma stem-like cells (GSCs). This rare tumorigenic cell subpopulation resists temozolomide, undergoes proneural-to-mesenchymal transition (PMT) to evade therapy, and drives recurrence. Through inference of transcriptional regulatory networks (TRNs) of patient-derived GSCs (PD-GSCs) at single-cell resolution, we demonstrate how the topology of transcription factor interaction networks drives distinct trajectories of cell-state transitions in PD-GSCs resistant or susceptible to cytotoxic drug treatment. By experimentally testing predictions based on TRN simulations, we show that drug treatment drives surviving PD-GSCs along a trajectory of intermediate states, exposing vulnerability to potentiated killing by siRNA or a second drug targeting treatment-induced transcriptional programs governing nongenetic cell plasticity. Our findings demonstrate an approach to uncover TRN topology and use it to rationally predict combinatorial treatments that disrupt acquired resistance in GBM.

Dimensional reduction techniques have long been used to visualize the structure and geometry of high dimensional data. However, most widely used techniques are difficult to interpret due to nonlinearities and opaque optimization processes. Here we present a specific graph based construction for dimensionally reducing continuous stochastic systems with multiplicative noise moving under the influence of a potential. To achieve this, we present a specific graph construction which generates the Fokker-Planck equation of the stochastic system in the continuum limit. The eigenvectors and eigenvalues of the normalized graph Laplacian are used as a basis for the dimensional reduction and yield a low dimensional representation of the dynamics which can be used for downstream analysis such as spectral clustering. We focus on the use case of single cell RNA sequencing data and show how current diffusion map implementations popular in the single cell literature fit into this framework.

Background: Unsupervised classification of brain cell types can be readily achieved through analysis of gene expression patterns in multidimensional space. However, the small transcriptional differences between microglia subtypes require more principled, robust, and reproducible approaches to expose and distinguish biological differences from noise.Method: Here we leverage Fokker‐Planck diffusion maps [1,2] to examine the dynamics of microglia subtypes by finding a low dimensional representation of the underlying stochastic dynamical system responsible for generating these transcriptional states. Clustering directly on this projection allows us to reduce noise, assign time scales to these substates, and generate a hierarchy of microglial cell states. Once these subtypes have been defined, we apply pseudobulk differential expression analysis to identify marker genes, and use those genes to build a boosted tree model to classify cells states. We then apply Shapley additive explanations (SHAP) to identify which genes are most important for distinguishing between cell states.Result: We observe a radial transition pattern emanating, with a subtype that appears homeostatic, inactivated, and poised for transition. This is consistent with the observation that microglia dynamics are sensitive to their microenvironment. Deconvolution of an independent bulk microglia dataset (n = 130) showed depletion in subtypes linked to protein misfolding and myelin phagocytosis, as well as enrichment in immunoreactive microglia in Alzheimer’s patients. SHAP reveals DUSP1, CD83, PLCG2, OLR1,CCL2, and JUN among the most important genes for differentiating into these AD‐associated clusters.Conclusion: Our approach enables us to reproducibly identify microglial subtypes across sets of single cell‐resolution high dimensional molecular profile data, identifying likely drivers of subtype differentiation. With several of our driver genes already associated with Alzheimer’s via GWAS and other analyses, our results will enable us to elucidate the role of these genes in determining microglial subtypes.References:[2] Nadler, B., Lafon, S., Coifman, R. R., & Kevrekidis, I. G. (2006). Diffusion maps, spectral clustering and reaction coordinates of dynamical systems. Applied and Computational Harmonic Analysis, 21(1), 113‐127.[2] Coifman, R. R., Kevrekidis, I. G., Lafon, S., Maggioni, M., & Nadler, B. (2008). Diffusion maps, reduction coordinates, and low dimensional representation of stochastic systems. Multiscale Modeling & Simulation, 7(2), 842‐864.

Recent studies at individual cell resolution have revealed phenotypic heterogeneity in nominally clonal tumor cell populations. The heterogeneity affects cell growth behaviors, which can result in departure from the idealized uniform exponential growth of the cell population. Here we measured the stochastic time courses of growth of an ensemble of populations of HL60 leukemia cells in cultures, starting with distinct initial cell numbers to capture a departure from the {uniform exponential growth model for the initial growth (``take-off'')}. Despite being derived from the same cell clone, we observed significant variations in the early growth patterns of individual cultures with statistically significant differences in growth dynamics, which could be explained by the presence of inter-converting subpopulations with different growth rates, and which could last for many generations. Based on the hypothesis of existence of multiple subpopulations, we developed a branching process model that was consistent with the experimental observations.