Biology subjects

Li, A. Z.

Publications and source records attributed to Li, A. Z..

6 recordsLinked to original sources

bioRxiv2026

A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics

Microorganisms are indispensable to terrestrial ecosystems, with their genomic material underpinning critical functions and applications across agriculture, biotechnology, and human health. Although genomic language models have advanced representation learning in DNA sequences, the extensive diversity of microorganisms and imbalanced taxonomic representation in the pretraining corpora pose challenges for effective microbial genomic sequence modeling. Here we present MicroGlot, a taxonomy-informed microbial DNA foundation model pretrained on 3.70 million sequences comprising 378.3 billion nucleotides across 99{,}700 species. MicroGlot encodes the hierarchical relations among taxa through hyperbolic embeddings, incorporating microbial taxonomic knowledge into a sparse mixture-of-experts architecture. Zero-shot evaluation of MicroGlot's layer embeddings demonstrates that the model's representations encode phenotypic traits and taxonomic identity. Comparison with a taxonomy-ablated variant trained under the same pretraining scheme shows that incorporating taxonomic knowledge consistently improves representation quality across the layers of MicroGlot. MicroGlot also combines optimized training techniques with efficient architectural components from modern large language models, achieving leading zero-shot performance across layers and competitive fine-tuning performance with low computational overhead. In a 1000-species set sampled from major cellular domains and viral realms, MicroGlot's routing fingerprints show greater agreement with taxonomic groups than tetranucleotide composition, reflecting taxonomically structured expert routing in multilingual modeling of microbial genomes. Overall, we show that MicroGlot serves as an efficient and effective DNA foundation model for microbial genomic analysis.

bioinformatics↗

bioRxiv2026

Stress Granule Coarsening Is a Pathological Inflection Point for Cardiac Electrophysiological Dysfunction

Arrhythmia risk rises early in many forms of cardiac stress, often before contractile failure is evident. Stressed cardiomyocytes accumulate biomolecular condensates, known as stress granules (SGs), whose contribution to this electrical vulnerability has been unclear. We mapped where SGs reside and followed their life cycle under acute and chronic oxidative stress across complementary model systems and assessed electrophysiological consequences with pharmacological tools targeting granule assembly, microtubule integrity, and calcium channel function. Under both stress regimes, SGs localized preferentially to z-lines and intercalated discs, marking these mechanically critical sites as hubs of condensate assembly. Merging of granules, referred to as coarsening, rather than initial formation, emerged as a pathological connection. Early granules were broadly cytoprotective, whereas progressive coarsening was accompanied by disruption of alpha-actinin and L-type calcium channel (Cav1.2) nanodomains and by shortening of action potential (AP) duration. Coarsened granules disorganized Cav1.2 nanodomains through a microtubule-dependent mechanism, and arresting coarsening with nocodazole preserved nanodomain integrity and restored AP morphology. The transition from nascent to coarsened SGs therefore represents a targetable inflection point, and limiting coarsening may prevent proarrhythmic remodeling during cardiac oxidative stress.

bioengineering↗

bioRxiv2025

Benchmarking computational methods for multi-omics biomarker discovery in cancer

Multi-omics profiling characterizes cancer biology and supports biomarker discovery for prognosis and therapy selection. Although numerous computational multi-omics biomarker identification methods have been proposed, their ability to identify clinically relevant biomarkers has not been systematically evaluated, leaving it unclear whether the resulting biomarker nominations are reliable for downstream validation. Here we systematically benchmark 20 representative statistical, machine learning and deep learning methods using curated gold-standard prognostic and therapeutic biomarkers across five real-world datasets. We evaluate performance in terms of both biomarker identification accuracy and stability. Overall, DeePathNet and Deep-KEGG achieve the best performance. Across methods, effective biomarker recovery is associated with the integration of biological knowledge, global feature interactions, multivariate feature attribution, and effective regularization. Analysis of omics type contributions reveals method- and modality-specific biases, highlighting the importance of broader omics integration. We further evaluate methods on simulated datasets to probe sensitivity with controlled signal and noise. By aggregating results from top-performing methods, we construct consensus biomarker panels that nominate candidates for potential investigations. Finally, we provide user-friendly interfaces to allow researchers to benchmark new methods against the 20 baselines or apply selected methods for biomarker identification on custom multi-omics datasets. Our benchmark is publicly available at https://github.com/athanzli/CancerMOBI-Bench.

bioinformatics↗

bioRxiv2024

NeuroTD: A Time-Frequency Based Multimodal Learning Approach to Analyze Time Delays in Neural Activities

Studying the temporal dynamics of neural activities is essential for understanding how neurons function. These dynamics often involve temporal delays between neurons that vary over time, revealing both their functions and how they interact within circuits. Recent techniques such as Neuropixels, depth electrodes, and Patch-seq enable time-series recordings of neural activity at various scales, ranging from single neurons to large populations. However, inferring such time-varying delays remains challenging due to noise, high sampling rates, and complex temporal patterns. To address these challenges, we developed NeuroTD, a novel computational approach based on sliding windows to align time-series datasets and infer time-varying delays. Particularly, NeuroTD integrates adaptive window-size tuning to obtain optimal and robust delay estimates. We first benchmarked NeuroTD in simulation studies, demonstrating its robustness and outperformance. Then we applied it to two emerging real-world datasets: (i) intracranial multi-channel electrophysiological recordings from depth electrodes across medial temporal lobe regions in humans, showing that hippocampal signals recorded via depth electrodes exhibited consistently significantly longer time delays than other regions during working memory tasks, and (ii) Patch-seq data in the mouse motor cortex, revealing intrinsic electro-physiological time-delays of excitatory neurons correlated with gene expression and highlighting pathways related to ion transport and neuronal excitability. Finally, NeuroTD is open-source and available at https://github.com/daifengwanglab/NeuroTD for general use.

bioinformatics↗

bioRxiv2023

Interrogating the function of bicistronic translational control elements to improve consistency of gene expression

Context independent gene expression is required for genetic circuits to maintain consistent and predicable behavior. Previous efforts to develop context independent translation have leveraged the helicase activity of translating ribosomes via bicistronic design translational control elements (BCDs) located within an efficiently translated leader peptide. We have developed a series of bicistronic translational control elements with strengths that span several orders of magnitude, maintain consistent expression levels across diverse sequence contexts, and are agnostic to common ligation sequences used in modular cloning systems. We have used this series of BCDs to investigate several features of this design, including the spacing of the start and stop codons, the nucleotide identity upstream of the start codon, and factors affecting translation of the leader peptide. To demonstrate the flexibility of this architecture and their value as a generic modular expression control cassette for synthetic biology, we have developed a set of robust BCDs for use in several Rhodococcus species.

synthetic biology↗

bioRxiv2020

A CAR RNA FISH assay reveals functional and spatial heterogeneity of chimeric antigen receptor T cells in tissue

Chimeric antigen receptor (CAR) T cells are engineered cells used in cancer therapy and are studied to treat infectious diseases. Trafficking and persistence of CAR T cells is an important requirement for efficacy to target cancer. Here, we describe a CAR RNA FISH histo-cytometry platform combined with a random reaction seed image analysis algorithm to quantitate spatial distribution and in vivo functional activity of a CAR T cell population at a single cell resolution for preclinical models. In situ, CAR T cell exhibited a heterogenous effector gene expression and this was related to the distance from tumor cells, allowing a quantitative assessment of the potential in vivo effectiveness. The platform offers the potential to study immune functions of genetically engineered cells in situ with their target cells in tissues with high statistical power and thus, can serve as an important tool for preclinical assessment of CAR T cell effectiveness. Brief summaryWe developed an imaging platform and analysis pipeline to study large populations of engineered cells at a single cell level in situ. One Sentence SummaryWe developed a CAR RNA FISH assay to study chimeric antigen receptor T cell trafficking and function in mouse tissue.

cancer biology↗