Biology subjects

Lee, L. J.

Publications and source records attributed to Lee, L. J..

13 recordsLinked to original sources

bioRxiv2026

Genos-m: a foundation model for human-associated microbial genomes

Human-associated microbial genomes encode extensive strain-level diversity and niche-specific gene repertoires that are critical to host health. However, these complex sequence features remain difficult to capture using general-purpose DNA foundation models, highlighting the need for dedicated representation learning tailored to the human microbiome. Here, we introduce Genos-m, an open-source foundation model for human-associated microbial genome representation. Genos-m was pretrained on approximately 1.2 trillion nucleotide tokens from a curated microbial genome corpus, including human-associated prokaryotic isolates, high-quality metagenome-assembled genomes (MAGs) and bacteriophages, supplemented with GTDB species-level representative genomes to broaden prokaryotic taxonomic breadth. The model uses a sparsely activated Mixture-of-Experts (MoE) Transformer architecture, with 4.7 billion total parameters, approximately 330 million activated parameters per forward pass and a maximum context length of one million base pairs. We evaluated frozen Genos-m representations across short-sequence and gene-level tasks, biosynthetic gene cluster (BGC)-based regional sequence tasks, whole-genome strain phenotype prediction, and zero-shot transfer on prokaryote-related RNAfitness assays. Across these benchmarks, Genos-m consistently ranked among the leading comparison models, with the best performance in five of eight gene-fitness regression tasks and in BGC type classification. Using sparse autoencoders, we identified sparse features in Genos-m hidden activations that aligned with annotated ORFs, intergenic regions, and tRNA and rRNA loci. In downstream applications, Genos-m-derived genome-informed species representations in-corporated into a human microbiome self-supervised learning model improved colorectal cancer (CRC)-control classification over conventional species-abundance random forest models. Genos-m also generated stable sample-level embeddings from as few as 10,000 metagenomic reads, retaining gut microbial community structure that distinguished geographic origin and aligned with enterotypes defined from full-depth taxonomic profiles. Together, these results support Genos-m as a reusable representation model for microbial genomes and metagenomes, with conclusions bounded by the reported datasets, task definitions and evaluation protocols. Genos-m model weights, inference code, and usage documentation are publicly available on GitHub (https://github.com/BGI-HangzhouAI/Genos-m) and Hugging-Face (https://huggingface.co/BGI-HangzhouAI/Genos-m).

bioinformatics↗

bioRxiv2025

Enigma: An Efficient Model for Deciphering Regulatory Genomics

Genomic sequence-to-function models have emerged as powerful tools for deciphering cis-regulatory grammar to advance our understanding of disease biology and guide therapeutic development. Recent advances have been driven by multi-task training of large transformer-based models on thousands of genome tracks. However, these performance gains have come at significant computational cost for both training and inference, hindering large-scale applications and slowing future model development. Here, rather than continuing to scale model size and add more training tracks, we focus on architectural efficiency and train on a substantially smaller, curated set of genome tracks. Our model Enigma achieves competitive performance with current state-of-the-art models at single-base resolution while substantially reducing computational cost. On zero-shot variant effect prediction benchmarks, Enigma outperforms the leading open-source model, the Borzoi ensemble, while using 10.9% of its compute and improving resolution from 32 bases to a single base. Compared to AlphaGenome, Enigma achieves 90.4 - 97.3% of its performance using 7.5% of its estimated compute. These improvements in efficiency can facilitate further development of models for regulatory genomics. We demonstrate this by fine-tuning Enigma on predicting three new molecular phenotypes -- ChIP-seq, RNA half-life, and translation efficiency -- achieving or exceeding the performance of state-of-the-art task-specific models. We are providing Enigma for non-commercial use to benefit the broader research field. [Formula]Code: https://github.com/deepgenomics/enigma

genomics↗

bioRxiv2025

FlashRNA: An Efficient Model for Regulatory Genomics

Transformer-based genomic sequence-to-function models effectively capture long-range genomic interactions but incur high computational costs due to the quadratic complexity of their self-attention layers. In this work, we introduce FlashRNA, which significantly improves computational and memory efficiency through FlashAttention, advancements in model architecture, and optimized training setup. FlashRNA achieves comparable or slightly improved predictive performance compared to similar sized Borzoi or Flashzoi models, notably without depending on pre-trained weights - a major limitation of Flashzoi. Remarkably, we trained FlashRNA from scratch in one day on a single GPU, significantly accelerating training and inference speed. These improvements can facilitate further developments in models for regulatory genomics by reducing computational cost. We demonstrate this in two downstream applications: 1) we train a large ensemble of 16 FlashRNA models and distill them into a single model to improve performance while maintaining efficiency, and 2) we fine-tune FlashRNA on three prediction tasks - ChIP-seq, RNA half-life, and translation efficiency - achieving performance matching or exceeding state-of-the-art task-specific models. Code: https://github.com/deepgenomics/flashrna

genomics↗

bioRxiv2024

Orthrus: Towards Evolutionary and Functional RNA Foundation Models

In the face of rapidly accumulating genomic data, our ability to accurately predict key mature RNA properties that underlie transcript function and regulation remains limited. Pre-trained genomic foundation models offer an avenue to adapt learned RNA representations to biological prediction tasks. However, existing genomic foundation models are trained using strategies borrowed from textual domains that do not leverage biological domain knowledge. Here, we introduce Orthrus, a Mamba-based mature RNA foundation model pre-trained using a novel self-supervised contrastive learning objective with biological augmentations. Orthrus is trained by maximizing embedding similarity between curated pairs of RNA transcripts, where pairs are formed from splice isoforms of 10 model organisms and transcripts from orthologous genes in 400+ mammalian species from the Zoonomia Project. This training objective results in a latent representation that clusters RNA sequences with functional and evolutionary similarities. We find that the generalized mature RNA isoform representations learned by Orthrus significantly outperform genomic foundation models on mRNA property prediction tasks, and requires only a fraction of fine-tuning data to do so. Finally, we show that Orthrus is capable of capturing divergent biological function of individual transcript isoforms.

genomics↗

bioRxiv2022

Engineering a Single Extracellular Vesicle Protein and RNA Assay (siEVPRA) via In Situ Fluorescence Microscopy in a UV Micropatterned Array

The physical and molecular heterogeneity of extracellular vesicles (EVs) confounds bulk biomarker characterization, thus encouraging the development of novel assays capable of profiling EVs at a single-vesicle resolution. Here, we present a single EV (siEV) protein and RNA assay (siEVPRA) to simultaneously detect proteins, messenger RNAs (mRNAs), and microRNAs (miRNAs) in siEVs. The siEVPRA consists of an array of microdomains embedded on a polyethylene glycol (PEG)-coated glass surface produced via UV photopatterning, functionalized with antibodies to target siEV subpopulations. Fluorescently labeled antibodies and RNA-targeting molecular beacons (MBs) were used to generate signals for proteins, mRNAs, and miRNAs on siEVs detected by total internal reflection fluorescence microscopy (TIRFM), outperforming the sensitivities of ELISA and PCR by three orders of magnitude. Using the siEVPRA, we analyzed EVs harvested from glioblastoma (GBM) cell lines and demonstrated vesicular heterogeneity in protein, mRNA, and miRNA expression through colocalization analyses, and validated the results by bulk RNA sequencing. We further demonstrated the clinical utility of the siEVPRA by detecting different mRNAs and miRNAs associated with GBM in patient samples. Together, these results indicate that the siEVPRA provides an effective platform to investigate the heterogeneity of proteins and RNAs in subpopulations of EVs.

bioengineering↗

bioRxiv2022

mTOR-regulated Mitochondrial Metabolism Limits Mycobacterium-induced Cytotoxicity

Necrosis of macrophages in the tuberculous granuloma represents a major pathogenic event in tuberculosis. Through a zebrafish forward genetic screen, we identified the mTOR kinase, a master regulator of metabolism, as an early host resistance factor in tuberculosis. We found that mTOR complex 1 protects macrophages from mycobacterium-induced death by enabling infection-induced increases in mitochondrial energy metabolism fueled by glycolysis. These metabolic adaptations are required to prevent mitochondrial damage and death caused by the secreted mycobacterial virulence determinant ESAT-6. Thus, the host can effectively counter this early critical mycobacterial virulence mechanism simply by regulating energy metabolism, thereby allowing pathogen-specific immune mechanisms time to develop. Our findings may explain why Mycobacterium tuberculosis, albeit humanitys most lethal pathogen, is successful in only a minority of infected individuals.

immunology↗

bioRxiv2021

A graph neural network approach to molecule carcinogenicity prediction

Molecular carcinogenicity is a preventable cause of cancer, however, most experimental testing of molecular compounds is an expensive and time consuming process, making high throughput experimental approaches infeasible. In recent years, there has been substantial progress in machine learning techniques for molecular property prediction. In this work, we propose a model for carcinogenicity prediction, CONCERTO, which uses a graph transformer in conjunction with a molecular fingerprint representation, trained on multi-round muta-genicity and carcinogenicity objectives. To train and validate CONCERTO, we augment the training dataset with more informative labels and utilize a larger external validation dataset. Extensive experiments demonstrate that our model yields results superior to alternate approaches for molecular carcinogenicity prediction.

bioinformatics↗

bioRxiv2021

An Immunogold Single Extracellular Vesicular RNA and Protein (AuSERP) Biochip to Predict Responses to Immunotherapy in Non-Small Cell Lung Cancer Patients

Conventional PD-L1 immunohistochemical tissue biopsies only predict 20~40% of non-small cell lung cancer (NSCLC) patients that will respond positively to anti-PD-1/PD-L1 immunotherapy. Herein, we present an immunogold biochip to quantify single extracellular vesicular RNA and protein (AuSERP) as a non-invasive alternative. With only 3 L of serum, PD-1/PD-L1 proteins on the surface of extracellular vesicles (EVs) and EV PD-1/PD-L1 messenger RNA (mRNA) cargo were detected at a single-vesicle resolution and exceeded the sensitivities of ELISA and qRT-PCR by 1000 times. By testing a cohort of 27 non-responding and 27 responding NSCLC patients, AuSERP indicated that the single-EV mRNA biomarkers surpass the single-EV protein biomarkers at predicting patient responses to immunotherapy. Dual single-EV PD-1/PD-L1 mRNA detection differentiated responders from non-responders with an accuracy of 72.2% and achieved an NSCLC diagnosis accuracy of 93.2%, suggesting the potential for AuSERP to provide enhanced immunotherapy predictions and cancer diagnoses within the clinical setting.

immunology↗

bioRxiv2021

IntroSpect: motif-guided immunopeptidome database building tool to improve the sensitivity of HLA binding peptide identification

Although database search tools originally developed for shotgun proteome have been widely used in immunopeptidomic mass spectrometry identifications, they have been reported to achieve undesirably low sensitivities and/or high false positive rates as a result of the hugely inflated search space caused by the lack of specific enzymic digestions in immunopeptidome. To overcome such a problem, we have developed a motif-guided immunopeptidome database building tool named IntroSpect, which is designed to first learn the peptide motifs from high confidence hits in the initial search and then build a targeted database for refined search. Evaluated on three representative HLA class I datasets, IntroSpect can improve the sensitivity by an average of 80% comparing to conventional searches with unspecific digestions while maintaining a very high accuracy ([~]96%) as confirmed by synthetic validation experiments. A distinct advantage of IntroSpect is that it does not depend on any external HLA data so that it performs equally well on both well-studied and poorly-studied HLA types, unlike a previously developed method SpectMHC. We have also designed IntroSpect to keep a global FDR that can be conveniently controlled, similar to conventional database search engines. Finally, we demonstrate the practical value of IntroSpect by discovering neoantigens from MS data directly. IntroSpect is freely available at https://github.com/BGI2016/IntroSpect.

bioinformatics↗

bioRxiv2020

HiCSR: a Hi-C super-resolution framework for producing highly realistic contact maps

MotivationHi-C data has enabled the genome-wide study of chromatin folding and architecture, and has led to important discoveries in the structure and function of chromatin conformation. Here, high resolution data plays a particularly important role as many chromatin substructures such as Topologically Associating Domains (TADs) and chromatin loops cannot be adequately studied with low resolution contact maps. However, the high sequencing costs associated with the generation of high resolution Hi-C data has become an experimental barrier. Data driven machine learning models, which allow low resolution Hi-C data to be computationally enhanced, offer a promising avenue to address this challenge. ResultsBy carefully examining the properties of Hi-C maps and integrating various recent advances in deep learning, we developed a Hi-C Super-Resolution (HiCSR) framework capable of accurately recovering the fine details, textures, and substructures found in high resolution contact maps. This was achieved using a novel loss function tailored to the Hi-C enhancement problem which optimizes for an adversarial loss from a Generative Adversarial Network (GAN), a feature reconstruction loss derived from the latent representation of a denoising autoencoder, and a pixel-wise loss. Not only can the resulting framework generate enhanced Hi-C maps more visually similar to the original high resolution maps, it also excels on a suite of reproducibility metrics produced by members of the ENCODE Consortium compared to existing approaches, including HiCPlus, HiCNN, hicGAN and DeepHiC. Finally, we demonstrate that HiCSR is capable of enhancing Hi-C data across sequencing depth, cell types, and species, recovering biologically significant contact domain boundaries. AvailabilityWe make our implementation available for download at: https://github.com/PSI-Lab/HiCSR Contactljlee@psi.toronto.edu Supplementary informationAvailable Online

genomics↗

bioRxiv2019

The co-stimulation of anti-CD28 and IL-2 enhances the sensitivity of ELISPOT assays for detection of neoantigen-specific T cells in PBMC

Neoantigen-based cancer immunotherapies hold the promise of being a truly personalized, effective treatment for diverse cancer types. ELISPOT assays, as a powerful experimental technique, can verify the existence of antigen specific T cells to support basic clinical research and monitor clinical trials. However, despite the high sensitivity of ELISPOT assays, detecting immune responses of neoantigen specific T cells in a patient or healthy donors PBMCs is still extremely difficult, since the frequency of these T cells can be very low. We developed a novel experimental method, by co-stimulation of T cells with anti-CD28 and IL-2 at the beginning of ELISPOT, to further increase the sensitivity of ELISPOT and mitigate the challenge introduced by low frequency T cells. Under the optimal concentration of 1g/mL for anti-CD28 and 1U/ml for IL-2, our new method can increase sensitivity by up to 5 folds comparing to a conventional ELISPOT, and outperforms other cytokine stimulation alternatives. To the best of our knowledge, this is the first report that the co-stimulation of anti-CD28 and IL-2 is able to significantly improve the sensitivity of ELISPOT assays, indicating that anti-CD28 and IL-2 signaling can act in synergy to lower the T cell activation threshold and trigger more neoantigen-specific T cells.

immunology↗

bioRxiv2018

cDeepbind: A context sensitive deep learning model of RNA-protein binding

MotivationDetermining RNA binding protein(RBP) binding specificity is crucial for understanding many cellular processes and genetic disorders. RBP binding is known to be affected by both the sequence and structure of RNAs. Deep learning can be used to learn generalizable representations of raw data and has improved state of the art in several fields such as image classification, speech recognition and even genomics. Previous work on RBP binding has either used shallow models that combine sequence and structure or deep models that use only the sequence. Here we combine both abilities by augmenting and refining the original Deepbind architecture to capture structural information and obtain significantly better performance.\n\nResultsWe propose two deep architectures, one a lightweight convolutional network for transcriptome wide inference and another a Long Short-Term Memory(LSTM) network that is suitable for small batches of data. We incorporate computationally predicted secondary structure features as input to our models and show its effectiveness in boosting prediction performance. Our models achieved significantly higher correlations on held out in-vitro test data compared to previous approaches, and generalise well to in-vivo CLIP-SEQ data achieving higher median AUCs than other approaches. We analysed the output from our model for VTS1 and CPO and provided intuition into its working. Our models confirmed known secondary structure preferences for some proteins as well as found new ones where secondary structure might play a role. We also demonstrated the strengths of our model compared to other approaches such as the ability to combine information from long distances along the input.\n\nAvailabilitySoftware and models are available at https://github.com/shreshthgandhi/cDeepbind\n\nContactljlee@psi.toronto.edu, frey@psi.toronto.edu

genomics↗

bioRxiv2018

Isw2 and Ino80 chromatin remodeling factors regulate chromatin, replication, and copy number at the yeast ribosomal DNA locus

In the budding yeast Saccharomyces cerevisiae, ribosomal RNA genes are encoded in a highly repetitive tandem array referred to as the ribosomal DNA (rDNA) locus. The yeast rDNA is the site of a diverse set of DNA-dependent processes, including transcription of ribosomal RNAs by RNA Polymerases I and III, transcription of non-coding RNAs by RNA Polymerase II, DNA replication initiation, replication fork blocking, and recombination-mediated regulation of rDNA repeat copy number. All of this takes place in the context of chromatin, but relatively little is known about the roles played by ATP-dependent chromatin remodeling factors at the yeast rDNA. In this work, we report that the Isw2 and Ino80 chromatin remodeling factors are targeted to this highly repetitive locus. We characterize for the first time their function in modifying local chromatin structure, finding that loss of these factors affects the occupancy of nucleosomes in the 35S ribosomal RNA gene and the positioning of nucleosomes flanking the ribosomal origin of replication. In addition, we report that Isw2 and Ino80 promote efficient firing of the ribosomal origin of replication and facilitate the regulated increase of rDNA repeat copy number. This work significantly expands our understanding of the importance of ATP-dependent chromatin remodeling for rDNA biology.\n\nAuthor SummaryTo satisfy high cellular demand for ribosomes, genomes contain many copies of the genes encoding the RNA components of ribosomes. In the budding yeast Saccharomyces cerevisiae, these ribosomal RNA genes are located in the \"ribosomal DNA locus\", a highly repetitive array that contains approximately 150 copies of the same unit, in contrast to the single copies that suffice for most genes. This repetitive quality creates unique regulatory needs. Chromatin structure, the packaging and organization of DNA, is a critical determinant of DNA-dependent processes throughout the genome. ATP-dependent chromatin remodeling factors are important regulators of chromatin structure, and yet relatively little is known about how members of this class of protein affect DNA organization or behavior at the rDNA. In this work, we show that the Isw2 and Ino80 chromatin remodeling factors regulate two features of chromatin structure at the rDNA, the occupancy and the positioning of nucleosomes. In addition, we find that these factors regulate two critical processes that function uniquely at this locus: DNA replication originating from within the rDNA array, and the regulated increase of rDNA repeat copy number.

molecular biology↗