Biology subjects

Mostafavi, S.

Publications and source records attributed to Mostafavi, S..

At least 19 recordsLinked to original sources

bioRxiv2026

A Gene-Program Architecture of Mouse T cells

We present immgenT-GP, a gene-program framework for resolving mouse T cell heterogeneity across the immgenT atlas. On ~ 800,000 T cells spanning lineages, organs, and immune challenges, we defined 200 reproducible gene programs, discovered by empirical Bayes matrix factorization approach and validated through a new deep learning approach, that capture major axes of T cell variation, including lineage identity, activation states and tissue location. Gene-program analysis complemented cluster-based annotation by decomposing T cell states into molecular modules, revealing quantitative, shared, modules not represented with discrete labels alone. Across tissues, gene programs reflected both tissue-imposed programs and changes in cluster composition. Integrating GP activity with cell-surface marker expression from the CITE-seq data, revealed that markers can report different programs depending on lineage and context. Together, immgenT-GP extends the atlas from a map of T cell states to a molecular reference of the programs that underlie them.

genomics↗

bioRxiv2026

A modality gap in personal-genome prediction by sequence-to-function models

Sequence-to-function (S2F) models trained on reference genomes have achieved strong performance on regulatory prediction and variant-effect benchmarks, yet they still struggle to predict inter-individual variation in gene expression from personal genomes. We evaluated AlphaGenome on personal genome prediction in two molecular modalities--gene expression and chromatin accessibility--and observed a striking dichotomy: AlphaGenome approaches the heritability ceiling for chromatin accessibility variation, but remains far below baseline for gene-expression variation, despite improving over Borzoi. Context truncation and fine-mapped QTL analyses indicate that accessibility is governed by local regulatory grammar captured by current architectures, whereas gene-expression variation requires long-range regulatory integration that remains challenging.

bioinformatics↗

bioRxiv2025

GAME: Genomic API for Model Evaluation

The rapid expansion of genomics datasets and the application of machine learning has produced sequence-to-activity genomics models with ever-expanding capabilities. However, benchmarking these models on practical applications has been challenging because individual projects evaluate their models in ad hoc ways, and there is substantial heterogeneity of both model architectures and benchmarking tasks. To address this challenge, we have created GAME, a system for large-scale, community-led standardized model benchmarking on user-defined evaluation tasks. We borrow concepts from the Application Programming Interface (API) paradigm to allow for seamless communication between pre-trained models and benchmarking tasks, ensuring consistent evaluation protocols. Because all models and benchmarks are inherently compatible in this framework, the continual addition of new models and new benchmarks is easy. We also developed a Matcher module powered by a large language model (LLM) to automate ambiguous task alignment between benchmarks and models. Containerization of these modules enhances reproducibility and facilitates the deployment of models and benchmarks across computing platforms. By focusing on predicting underlying biochemical phenomena (e.g. gene expression, open chromatin, DNA binding), we ensure that tasks remain technology-independent. We provide examples of benchmarks and models implementing this framework, and anticipate that the community will contribute their own, leading to an ever-expanding and evolving set of models and evaluation tasks. This resource will accelerate genomics research by illuminating the best models for a given task, motivating novel functional genomic benchmarks, and providing a more nuanced understanding of model abilities.

bioinformatics↗

bioRxiv2025

Deep genomic models of allele-specific measurements

Allele-specific quantification of sequencing data allows for a causal investigation of how DNA sequence variations influence cis gene regulation. Current methods for analyzing allele-specific measurements for causal analysis rely on statistical associations between genetic variation across individuals and allelic imbalance. Instead, we propose DeepAllele, a novel deep learning sequence-to-function model using paired allele-specific input, designed to learn sequence features that predict subtle changes in gene regulation between alleles. Our approach is suited for datasets with unambiguous phasing, such as F1 hybrids and other controlled genetic crosses, or long-read sequencing technologies used in Fiber-seq, in which reads can be assigned to complete allele sequences. We apply our framework to allele-specific measurements in immune cells from F1 hybrid mice, and show that the models additionally learned cis-regulatory grammar aligns with known biological mechanisms across a significantly larger number of genomic regions compared to baseline models. In summary, our work presents a computational framework to leverage genetic variation to uncover functionally-relevant regulatory motifs, enhancing causal discovery in genomics.

bioinformatics↗

bioRxiv2025

Investigating Data Size, Sequence Diversity, and Model Complexity in MPRA-based Sequence-to-Function Prediction

We created the MPRA Dataset Collection (MDC), a curated resource of MPRA data from 12 studies comprising over 150 million labeled DNA subsequences. These datasets include both random and natural genomic sequences paired with diverse functional outputs such as gene expression and splicing efficiency. Using this collection, we build sequence-to-function (S2F) predictive models of regulatory elements and analyze these models to uncover insights into the relationship between training data requirements, experimental design, and model generalizability. By offering a high-quality, machine learning-ready repository, MDC accelerates the development of robust computational tools for deciphering the mechanisms of gene regulation.

bioinformatics↗

bioRxiv2025

A scalable approach to investigating sequence-to-expression prediction from personal genomes

Sequence-to-function (S2F) models hold the promise of evaluating arbitrary DNA sequences, providing a powerful framework for linking genotype to phenotype. Yet, despite strong performance across genomic loci, these models often struggle to capture inter-individual variation in gene expression. To address this, we propose personal genome training--training models to make genotype-specific predictions at a single locus. We introduce SAGE-net, a scalable framework and software package for training and evaluating S2F models using personal genomes. Using SAGE-net, we systematically explore model architectures and training regimes, showing that personal genome training improves gene expression prediction accuracy for held-out individuals. However, performance gains arise primarily from identifying predictive variants, rather than learning a cis-regulatory grammar that generalizes across loci. This lack of generalization persists across a wide range of hyperparameters. In contrast, when applied to DNA methylation (DNAm), personal genome training enables improved generalization to unseen individuals in unseen genomic regions. This suggests that S2F models may more readily capture the sequence-level determinants of inter-individual variation in epigenomic traits. These findings highlight the need for further exploration to unlock the full potential of S2F models in decoding the regulatory grammar of personal genomes. Scalable software and infrastructure development will be critical to this progress.

genomics↗

bioRxiv2025

Structural and Functional Versatility of the Amyloidogenic Antimicrobial Peptide Citropin 1.3

Citropin 1.3 is an antimicrobial peptide produced by the amphibian Litoria citropa (Southern bell frog), which self-aggregates into distinct fibrillar structures, however, the function of the fibrils remains unclear and largely unexplored. In this study, the structural and functional properties of citropin 1.3 were investigated using cryogenic electron microscopy and fluorescence microscopy in the presence of membrane and cell models, and with X-ray crystallography. Canonical amyloids, multilayered nanotubes, and a novel mixed fibril were observed. Experiments with negatively charged giant unilamellar vesicles revealed that the peptide facilitates membrane fusion while simultaneously undergoing phase separation in the presence of phospholipids. In presence of mammalian cells, citropin 1.3 permeabilizes membranes, leading to cell death, and over time, colocalizes with genetic material. Overall, this work provides new insights into the structural dynamics of the amyloidogenic antimicrobial peptide citropin 1.3 and its interactions with different systems.

molecular biology↗

bioRxiv2025

Refining the cis-regulatory grammar learned by sequence-to-activity models by increasing model resolution

Decoding the cis-regulatory syntax that controls gene expression is essential for improving our understanding of cell differentiation and disease. To identify regulatory motifs and their regulatory syntax, deep learning based sequence-to-activity (S2A) models learn transcription factor binding motifs and their combinations from DNA sequence by modeling measured chromatin accessibility. Previously, we developed AI-TAC, a S2A model that predicts chromatin accessibility across various immune cell types in multi-task fashion, effectively decoding the regulatory syntax underlying immune cell differentiation. While ATAC-seq is commonly used to measure regional accessibility, it also provides high-resolution profiles, the distribution of Tn5 insertion sites, that offer additional insights into the precise location and strength of TF binding sites. Here we demonstrate that modeling ATAC-seq profiles alongside accessibility consistently improves predictions of differential chromatin accessibility across cell types. Moreover, we also find that multi-task learning across related immune cell types consistently outperforms single-task models. To understand what additional information bpAI-TAC learns from ATAC-seq profiles, we systematically compare sequence attributions from models trained with and without ATAC-seq profiles. We identify novel motifs with strong effect sizes that emerge only when profile data is included. Our findings suggest that modeling ATAC-seq at base-pair resolution enables the model to learn a more nuanced and sensitive representation of the cis-regulatory syntax driving immune cell-specific chromatin landscapes.

genomics↗

bioRxiv2024

Structural organization of p62 filaments and the cellular ultrastructure of calcium-rich p62-enwrapped lipid droplet cargo

The selective autophagy receptor p62/SQSTM1 (from hereon p62) is known to form higher-order filaments in vitro and to undergo liquid-liquid phase separation when mixed with poly-ubiquitin. We determined the full-length cryo-EM structure of p62 and elucidated a structured double helical filament scaffold composed of the PB1-domain associated with the flexible C-terminal part residing in the lumen and the solvent-accessible major groove. At different pH values and upon binding to soluble LC3, LC3-conjugated membranes and poly-ubiquitin, we observed p62 filament re-arrangements in the form of structural unwinding, disassembly, lateral association and bundling, respectively. In the cellular environment, under conditions of ATG5 knockdown leading to stalled autophagy, we imaged high-contrast layers consisting of p62 oligomers enwrapping lipid droplets by cryogenic electron tomography in cellulo, which we identified as calcium phosphate by compositional spectroscopy analysis. Together, we visualized the cellular ultrastructure of p62 oligomers with high calcium content as a potential intermediate step of autophagy initiation.

biophysics↗

bioRxiv2024

Mitochondrial complex I deficiency occurs in skeletal muscle of a subgroup of individuals with Parkinson's disease

Widespread neuronal complex I (CI) deficiency was recently reported to be a characteristic in a subgroup of individuals with idiopathic Parkinsons disease (PD). Here, we sought to determine whether a CI deficient subgroup could be discerned using clinically accessible muscle biopsy. Vastus lateralis needle biopsies were collected from 83 individuals with PD and 29 neurologically healthy controls and analyzed by immunohistochemistry for complexes I and IV, cytochrome c oxidase/succinate dehydrogenase (COX/SDH) histochemistry, and spectrophotometric activity assays of complexes I-IV. Mitochondrial DNA (mtDNA) copy number, deletions, and point variation were analyzed in single muscle fibers and bulk biopsy samples. PD muscle exhibited reduced CI activity at the group level, with 9% of cases falling below two standard deviations of the control group. This deficiency was not associated with mtDNA abnormalities. Our findings support the existence of a PD subpopulation characterized by CI pathology and suggest that stratification by extra-neural mitochondrial dysfunction may be informative for selecting individuals for clinical trials.

neuroscience↗

bioRxiv2024

Structural basis of substrate transport and drug recognition by the human thiamine transporter SLC19A3

Thiamine (vitamin B1) functions as an essential coenzyme in cells. Humans and other mammals cannot synthesise this vitamin de novo and thus have to take it up from their diet. Eventually, every cell needs to import thiamine across its plasma membrane which is mainly mediated by two specific thiamine transporters SLC19A2 and SLC19A3. Loss of function mutations in either of these transporters leads to detrimental, life-threatening metabolic disorders. SLC19A3 is furthermore a major site of drug interactions. Many medications, including antidepressants, antibiotics and chemotherapeutics are known to inhibit this transporter, with potentially fatal consequences for patients. Despite a thorough functional characterisation over the past two decades, the structural basis of its transport mechanism and drug interactions has remained elusive. Here, we report eight cryo-electron microscopy (cryo-EM) structures of the human thiamine transporter SLC19A3 in complex with various ligands. Conformation-specific nanobodies enabled us to capture different states of SLC19A3s transport cycle, revealing the molecular details of thiamine recognition and transport. We identified nine novel drug interactions of SLC19A3 and determined structures of the transporter in complex with the inhibitors fedratinib, hydroxychloroquine, amprolium and amitriptyline. These data allow us to develop an understanding of the transport mechanism and ligand recognition of SLC19A3.

biochemistry↗

bioRxiv2023

Quick and effective approximation of in silico saturation mutagenesis experiments with first-order Taylor expansion

To understand the decision process of genomic sequence-to-function models, various explainable AI algorithms have been proposed. These methods determine the importance of each nucleotide in a given input sequence to the models predictions, and enable discovery of cis regulatory motif grammar for gene regulation. The most commonly applied method is in silico saturation mutagenesis (ISM) because its per-nucleotide importance scores can be intuitively understood as the computational counterpart to in vivo saturation mutagenesis experiments. While ISM is highly interpretable, it is computationally challenging to perform, because it requires computing three forward passes for every nucleotide in the given input sequence; these computations add up when analyzing a large number of sequences, and become prohibitive as the length of the input sequences and size of the model grows. Here, we show how to use the first-order Taylor approximation to compute ISM, which reduces its computation cost to a single forward pass for an input sequence. We use our theoretical derivation to connect ISM with the gradient of the model and show how this approximation is related to a recently suggested correction of the models gradients for genomic sequence analysis. We show that the Taylor ISM (TISM) approximation is robust across different model ablations, random initializations, training parameters, and data set sizes.

bioinformatics↗

bioRxiv2023

Mitochondrial complex I deficiency stratifies idiopathic Parkinson's disease

Idiopathic Parkinsons disease (iPD) is believed to have a heterogeneous pathophysiology, but molecular disease subtypes have not been identified. Here, we show that iPD can be stratified according to the severity of neuronal respiratory complex I (CI) deficiency, and identify two emerging disease subtypes with distinct molecular and clinical profiles. The CI deficient (CI-PD) subtype accounts for approximately a fourth of all cases, and is characterized by anatomically widespread neuronal CI deficiency, a distinct cell type-specific gene expression profile, increased load of neuronal mtDNA deletions, and a predilection for non-tremor dominant motor phenotypes. In contrast, the non-CI deficient (nCI-PD) subtype exhibits no evidence of mitochondrial impairment outside the dopaminergic substantia nigra and has a predilection for a tremor dominant phenotype. These findings constitute a step towards resolving the biological heterogeneity of iPD with implications for both mechanistic understanding and treatment strategies.

neuroscience↗

bioRxiv2023

Impact of random 50-base sequences inserted into an intron on splicing in Saccharomyces cerevisiae

Intron splicing is a key regulatory step in gene expression in eukaryotes. Three sequence elements required for splicing - 5 and 3 splice sites and a branch point - are especially well- characterized in Saccharomyces cerevisiae, but our understanding of additional intron features that impact splicing in this organism is incomplete, due largely to its small number of introns. To overcome this limitation, we constructed a library in S. cerevisiae of random 50-nucleotide elements (N50) individually inserted into the intron of a reporter gene and quantified canonical splicing and the use of cryptic splice sites by sequencing analysis. More than 70% of approximately 140,000 N50 elements reduced splicing by at least 20% compared to the intron control. N50 features, including higher GC content, presence of GU repeats and stronger predicted secondary structure of its pre-mRNA, correlated with reduced splicing efficiency. A likely basis for the reduced splicing of such a large proportion of variants is the formation of RNA structures that pair N50 bases - such as the GU repeats - with other bases specifically within the reporter pre-mRNA analyzed. However, neither convolutional neural network nor linear models were able to explain more than a small fraction of the variance in splicing efficiency across the library, suggesting that complex non-linear interactions in RNA structures are not accurately captured by RNA structure prediction methods given the limited number of variants. Our results imply that the specific context of a pre-mRNA may determine the bases allowable in an intron to prevent secondary structures that reduce splicing.

synthetic biology↗

bioRxiv2023

How far are we from personalized gene expression prediction using sequence-to-expression deep neural networks?

Introductory Paragraph Deep learning methods have recently become the state-of-the-art in a variety of regulatory genomic tasks1-6 including the prediction of gene expression from genomic DNA. As such, these methods promise to serve as important tools in interpreting the full spectrum of genetic variation observed in personal genomes. Previous evaluation strategies have assessed their predictions of gene expression across genomic regions, however, systematic benchmarking is lacking to assess their predictions across individuals, which would directly evaluates their utility as personal DNA interpreters. We used paired Whole Genome Sequencing and gene expression from 839 individuals in the ROSMAP study7 to evaluate the ability of current methods to predict gene expression variation across individuals at varied loci. Our approach identifies a limitation of current methods to correctly predict the direction of variant effects. We show that this limitation stems from insufficiently learnt sequence motif grammar, and suggest new model training strategies to improve performance.

bioinformatics↗

bioRxiv2023

Gemini: Memory-efficient integration of hundreds of gene networks with high-order pooling

MotivationThe exponential growth of genomic sequencing data has created ever-expanding repositories of gene networks. Unsupervised network integration methods are critical to learn informative representations for each gene, which are later used as features for downstream applications. However, these network integration methods must be scalable to account for the increasing number of networks and robust to an uneven distribution of network types within hundreds of gene networks. ResultsTo address these needs, we present Gemini, a novel network integration method that uses memory-efficient high-order pooling to represent and weight each network according to its uniqueness. Gemini then mitigates the uneven distribution through mixing up existing networks to create many new networks. We find that Gemini leads to more than a 10% improvement in F1 score, 14% improvement in micro-AUPRC, and 71% improvement in macro-AURPC for protein function prediction by integrating hundreds of networks from BioGRID, and that Geminis performance significantly improves when more networks are added to the input network collection, while the comparison approachs performance deteriorates. Gemini thereby enables memory-efficient and informative network integration for large gene networks, and can be used to massively integrate and analyze networks in other domains. AvailabilityGemini can be accessed at: https://github.com/MinxZ/Gemini. Contactaddiewc@cs.washington.edu, swang@cs.washington.edu

bioinformatics↗

bioRxiv2022

Cell-subtype specific effects of genetic variation in the aging and Alzheimer cortex

The relationship between genetic variation and gene expression in individual brain cell types and subtypes has remained elusive. Here, we generated single-nucleus RNA sequencing data from the dorsolateral prefrontal cortex of 424 individuals of advanced age; analyzing 1.5 million nuclear transcriptomes, we assessed the effect of genetic variants on RNA expression in cis (cis-eQTL) for 7 cell types and 81 cell subtypes. This effort identified 10,004 eGenes at the cell type level and 8,138 eGenes at the cell subtype level. Many eGenes are only detected within cell subtypes. A new variant influences APOE expression only in microglia and is associated with greater cerebral amyloid angiopathy but not Alzheimer pathology, accounting for the effect of APOE{varepsilon}4, providing mechanistic insights into both pathologies. While eQTLs are readily detected, only a TMEM106B variant robustly affects the proportion of cell subtypes. Integration of these results with GWAS highlighted the targeted cell type and likely causal gene within susceptibility loci for Alzheimers, Parkinsons, schizophrenia, and educational attainment.

genomics↗

bioRxiv2022

ExplaiNN: interpretable and transparent neural networks for genomics

Sequence-based deep learning models, particularly convolutional neural networks (CNNs), have shown superior performance on a wide range of genomic tasks. A key limitation of these models is the lack of interpretability, slowing down their adoption by the genomics community. Current approaches to model interpretation do not readily reveal how a model makes predictions, can be computationally intensive, and depend on the implemented architecture. Here, we introduce ExplaiNN, an adaptation of neural additive models[1] for genomic tasks wherein predictions are computed as a linear combination of multiple independent CNNs, each consisting of a single convolutional filter and fully connected layers. This approach brings together the expressiveness of CNNs with the interpretability of linear models, providing global (cell state level) as well as local (individual sequence level) biological insights into the data. We use ExplaiNN to predict transcription factor (TF) binding and chromatin accessibility states, demonstrating performance levels comparable to state-of-the-art methods, while providing a transparent view of the models predictions in a straightforward manner. Applied to de novo motif discovery, ExplaiNN identifies equivalent motifs to those obtained from specialized algorithms across a range of datasets. Finally, we present ExplaiNN as a plug-and-play platform in which pretrained TF binding models and annotated position weight matrices from reference databases can be easily combined. We expect that ExplaiNN will accelerate the adoption of deep learning by biological domain experts in their daily genomic sequence analyses.

genomics↗