Biology subjects

Gagneur, J.

Publications and source records attributed to Gagneur, J..

At least 19 recordsLinked to original sources

bioRxiv2026

Scooby-2: fine-tuning AlphaGenome for single-cell genomic track prediction

Sequence-to-function models predict genomic tracks from DNA sequence. AlphaGenome has been shown to outperform previous models, including Borzoi, on most track-prediction and variant-effect benchmarks. However, like its predecessors, it predicts tracks only for predefined, mostly bulk-derived cell types and tissues. While Scooby has extended Borzoi to single-cell resolution, a single-cell adaptation of AlphaGenome is still lacking. We present Scooby-2, which combines AlphaGenome with a cell-specific decoder to predict single-cell RNA-seq coverage and chromatin-accessibility genomic tracks from DNA sequence. The package provides the implementation, the parameter-efficient fine-tuning setup, the training pipeline and inference scripts. When evaluated on the NeurIPS bone marrow single-cell dataset, Scooby-2 achieved a Pearson correlation of 0.56 when predicting cell-type-specific gene expression differences, compared with 0.53 for Scooby. When using the distilled AlphaGenome backbone, variant effect prediction concordance of fine-mapped eQTLs from the OneK1K cohort improved from 0.37 to 0.47 Spearman's rank correlation across all associations, an improvement consistent with that of AlphaGenome over Borzoi itself.

bioinformatics↗

bioRxiv2026

PepPr: De Novo peptide sequencing with peptide priors

Mass spectrometry-based peptide sequencing underpins proteomics. Database search confines to predefined candidates, whereas de novo peptide sequencing (DNPS) struggles with incomplete spectra. Existing hybrid methods use databases only for post hoc correction or retrieve previously identified spectra to guide decoding. We instead combine DNPS with a peptide prior: a language model fitted to a reference proteome, tolerating variants yet uncertain on strongly divergent sequences. Relative to prior-free DNPS, our implementation, PepPr, improved reference peptide identification and improved rather than suppressed variant identification. Using a related proteome as prior, it matched or outperformed homology-driven database search, and its gain over prior-free DNPS grew with relatedness. In antibody sequencing, it improved recovery of peptides spanning the variable-constant boundary, deamidation calls, and assembly, while preserving recall of hypervariable regions inaccessible to database search. Peptide priors thus bring reference knowledge into DNPS without limiting discovery.

bioinformatics↗

bioRxiv2026

Phenotypes of ultra-rare variant carriers benchmark variant effect scores

Human genetics needs benchmarks for evaluating variant scores against observed phenotypic consequences. We therefore present UKBBGym, which evaluates coding and non-coding scores against plasma protein abundance and quantitative traits observed in carriers of ultra-rare variants in the UK Biobank. This design avoids ascertainment biases of clinical labels, potential physiological mismatch of experimental assays, and confounding pertaining to common-variant associations. UKBBGym recapitulates relative performance on pathogenicity prediction while enabling comparison across variant classes and molecular mechanisms. This shows that scores designed for coding and splicing variants correlate more strongly with protein abundance and quantitative traits than scores modelling transcriptional regulation. Moreover, experimental assays do not consistently outperform computational scores for predicting missense effects. Comparing captured to detectable variance indicates scope for improvement, particularly for indels and loss-of-function variants. Finally, for coding variants, UKBBGym can be reproduced from public summary statistics, providing an accessible population-phenotype benchmark complementary to clinical labels and experimental assays.

genomics↗

bioRxiv2026

PARNET: A CLIP-SEQ-BASED FOUNDATION MODEL FOR RNA SEQUENCE REPRESENTATION LEARNING

RNA-binding proteins (RBPs) orchestrate a complex combinatorial regulatory "code" that governs RNA splicing, stability, localization, and translation. Learning the relationship between RNA sequences and these processes is a central challenge in genomics. Foundation models, notably RNA language models, have emerged as the dominant approach, learning general-purpose representations from unlabeled sequence at scale. While RNA language models have demonstrated impressive performance across a broad range of downstream tasks, they generally learn from sequence reconstruction objectives alone, lacking direct connections to the regulatory principles that govern RNA function. Here we introduce Parnet, an RNA foundation model trained directly and exclusively on experimental CLIP-seq data. Parnet is a multi-task foundation model trained end-to-end on 223 eCLIP-seq experiments spanning 150 RBPs to predict base-resolution RBP binding profiles directly from RNA sequence. This CLIP-seq pretraining strategy departs fundamentally from the masked-language-modeling paradigm, anchoring learned RNA representations directly in measured protein-RNA interactions rather than sequence statistics. Parnet substantially outperforms its single-task predecessor RBPNet in binding profile and motif recovery, generalizes to unseen cell types and iCLIP data, and recapitulates position-dependent splicing regulation. Frozen Parnet embeddings, without task-specific fine-tuning, match or exceed the performance of both task-specific tools, as well as larger self-supervised RNA and genomic language models across diverse downstream tasks, including RNA biotype classification, lncRNA chromatin localization, translational efficiency, splice-site recognition, intron retention, and non-coding variant effect prediction. Importantly, Parnet remains mechanistically interpretable, tracing predictions back to the specific RBPs and motifs that drive them. These results establish the RBP interactome as a compact, functionally sufficient, and interpretable basis for foundation model pretraining in RNA biology.

bioinformatics↗

bioRxiv2026

Ubiquitin selective ribosome profiling reveals systematic principles of co-translational quality control

Protein biogenesis is a stress- and error-sensitive process that can lead to nascent protein misfolding and aggregation, challenging cellular proteostasis.1 Co-translational ubiquitination (CTU) is a critical surveillance mechanism,2,3 yet its regulatory principles remain unclear due to limited known substrates and the lack of translatome-wide methods to query CTU. Here, we introduce UbSeRP, an approach that enables ubiquitin linkage-specific, and translatome-wide mapping of CTU. Focusing on ribosome-associated quality control (RQC), we expand the known endogenous RQC substrates in S. cerevisiae from few4-7 to thousands, reveal that only a subset of all identified disomes undergoes RQC, and uncover biophysical features of nascent chains that predict RQC engagement. We also identify widespread RQC-independent ubiquitination of nascent proteins, implicating broader roles of CTU in protein complex assembly. In a chronological aging model, we show that aging remodels translation and diminishes RQC engagement, favoring RQC-independent proteasomal pathways. Our findings provide systematic insight into the determinants and adaptability of CTU in maintaining proteostasis under changing physiological conditions.

biochemistry↗

bioRxiv2026

FENNEC: Fine-Tuned Ensemble Neural Networks Accelerate Chemically Modified siRNA Design and Screening

Small interfering RNAs (siRNAs) are a clinically validated therapeutic modality, yet designing potent chemically modified siRNAs remains a costly and iterative process, limited by scarce public data. Computational prediction of siRNA efficacy is therefore essential for rational design and accelerated preclinical development. However, despite the critical role of chemical modifications in therapeutic performance, current state-of-the-art machine learning methods either are not designed to model the chemical diversity of therapeutic siRNAs, or exhibit poor generalization performance. Here, we present FENNEC (Fine-Tuned Ensemble of Neural Networks for siRNA Efficiency Characterization), a machine-learning framework for predicting siRNA activity across chemically diverse design spaces. To support this effort, we curated the largest patent-derived dataset to date of chemically modified siRNAs from 42 patents using OCR-based table extraction and stringent filtering. FENNEC combines temporal convolutional networks with thermodynamic descriptors, experimental covariates, and embeddings from RNA foundation models to capture both local chemical determinants and broader target-context information. Importantly, we show that language-model-derived embeddings provide meaningful higher-order representations of target transcripts, particularly in data-scarce settings. FENNEC achieved robust predictive performance across both gene-level and scaffold-level validation settings, with additional experimental validation on a novel AHSA1-targeting dataset further supporting its generalizability across chemically modified siRNAs. In benchmarking, FENNEC outperformed classical machine-learning and state-of-the-art deep learning models, demonstrating generalization to unseen chemistry. Model interpretation recovered established design principles, including position-specific effects of glycol nucleic acid, 2'-fluoro modifications, and phosphorothioate backbones. Furthermore, in silico perturbation analyses suggest that FENNEC can serve not only as a predictive model, but also as an oracle for the design and optimization of chemically modified siRNAs. Together, our work addresses a key gap in the field by enabling chemically aware deep learning for siRNA design, supported by a large and diverse collection of chemically modified siRNA measurements.

bioinformatics↗

bioRxiv2025

Modanovo: A Unified Model for Post-Translational Modification-Aware de Novo Sequencing Using Experimental Spectra from In Vivo and Synthetic Peptides

Post-translational modifications (PTMs) play a central role in cellular regulation and are implicated in numerous diseases. Database searching remains the standard for identifying modified peptides from tandem mass spectra, but is hindered by the combinatorial expansion of modification types and sites. De novo peptide sequencing offers an attractive alternative, yet existing methods remain limited to unmodified peptides or a narrow set of PTMs. Here, we curated a large dataset of spectra from endogenous and synthetic peptides from ProteomeTools spanning 19 biologically relevant amino acid-PTM combinations, covering phosphorylation, acetylation, and ubiquitination. We used this dataset to develop Modanovo, an extension of the Casanovo transformer architecture for de novo peptide sequencing. Modanovo achieved robust performance across these amino acid-PTM combinations (median area under the precision-coverage curve 0.92), while maintaining performance on unmodified peptides (0.93), nearly identical to Casanovo (0.94). The model outperformed {pi}-PrimeNovo-PTM and showed increased precision and complementarity to the database search tool MSFragger. Robustness was confirmed across independent datasets, particularly at peptide lengths frequently represented in the curated dataset. Applied to a phosphoproteomics dataset from monkeypox virus-infected cells, Modanovo recovered numerous confident peptides not reported by database search, including new viral phosphosites supported by spectral evidence, thereby demonstrating its complementarity to database-driven identification approaches. These results establish Modanovo as a broadly applicable model for comprehensive de novo sequencing of both modified and unmodified peptides.

bioinformatics↗

bioRxiv2025

Aberrant splicing prediction during human organ development

Developmental disorders constitute a major class of genetic diseases, yet tools to identify splicing-disruptive variants during development are lacking. To address this need we extended the AbSplice framework to incorporate splicing dynamics across developmental stages. Moreover, we introduced several improvements including a refined ground truth from the aberrant splicing caller FRASER2, a continuous representation of splice site usage, and integration of the rich set of predictions from the sequence-based model Pangolin. These advances improve the performance of the original model at predicting aberrant splicing events and enable predictions across embryonic, childhood, and adult tissues. Genome-wide scores for all single-nucleotide variants and a web interface to score indels are available to facilitate the exploration of the predictions. Our genome-wide predictions reveal a class of variants with splicing-disruptive effects confined to early development, particularly abundant in the brain (>18,000 variants). These variants are enriched in loss-of-function intolerant genes and neurodevelopmental disorder genes. Within the rare-disease cohort Solve-RD high-impact brain-specific predictions are significantly enriched in individuals affected by neurodevelopmental disorders (NDD) in NDD-linked genes. Furthermore, the predicted developmental timing of the splicing disruption correlates with the clinical age of onset. Analysis of genomes of individuals with a suspected Mendelian disorder from Genomics England identified 26 unique variants in disease-linked genes, with stronger predicted effects during development than in adulthood, including a candidate new diagnosis in the gene FGFR1. Altogether, these results improve the accuracy of splice-disruptive variant prediction and provide tissue and developmental context to aid interpretation in rare disease diagnostics.

genomics↗

bioRxiv2025

PROTRIDER: Protein abundance outlier detection from mass spectrometry-based proteomics data with a conditional autoencoder

Structured abstractO_ST_ABSMotivationC_ST_ABSDetection of gene regulatory aberrations enhances our ability to interpret the impact of inherited and acquired genetic variation for rare disease diagnostics and tumor characterization. While numerous methods for calling RNA expression outliers from RNA-sequencing data have been proposed, the establishment of protein expression outliers from mass spectrometry data is lacking. ResultsHere, we propose and assess various modeling approaches to call protein expression outliers across three datasets from rare disease diagnostics and oncology. We use as independent evidence the enrichment for outlier calls in matched RNA-seq samples and the enrichment for rare variants likely disrupting protein expression. We show that controlling for hidden confounders and technical covariates, while simultaneously modeling the occurrence of missing values, is largely beneficial and can be achieved using conditional autoencoders. Moreover, we find that the differences between experimental and fitted log-transformed intensities by such models exhibit heavy tails that are poorly captured with the Gaussian distribution and report stronger statistical calibration when instead using the Students t-distribution. Our resulting method, PROTRIDER, outperformed baseline approaches based on raw log-intensities Z-scores or on differential expression analysis with limma. The application of PROTRIDER reveals significant enrichments of AlphaMissense pathogenic variants in protein expression outliers. Overall, PROTRIDER provides a method to confidently identify aberrantly expressed proteins applicable to rare disease diagnostics and cancer proteomics. Availability and ImplementationPROTRIDER is freely available at github.com/gagneurlab/PROTRIDER and also available on Zenodo under the DOI zenodo.15569781. ContactJulien Gagneur: gagneur at in.tum.de

bioinformatics↗

bioRxiv2024

Flashzoi: An enhanced Borzoi model for accelerated genomic analysis

Accurately predicting how DNA sequence drives gene regulation and how genetic variants alter gene expression is a central challenge in genomics. Borzoi, which models over ten thousand genomic assays including RNA-seq coverage from over half a megabase of sequence context alone promises to become an important foundation model in regulatory genomics, both for massively annotating variants and for further model development. However, its reliance on handcrafted, relative positional encodings within the transformer architecture limits its computational efficiency. Here we present Flashzoi, an enhanced Borzoi model that leverages rotary positional encodings and FlashAttention-2. This achieves over 3-fold faster training and inference and up to 2.4-fold reduced memory usage, while maintaining or improving accuracy in modeling various genomic assays including RNA-seq coverage, predicting variant effects, and enhancer-promoter linking. Flashzois improved efficiency facilitates large-scale genomic analyses and opens avenues for exploring more complex regulatory mechanisms and modeling. Availability and implementationThe Flashzoi model architecture is part of the MIT-licensed borzoi-pytorch package, can be found at https://github.com/johahi/borzoi-pytorch and installed via pip. Model weights for all four Flashzoi and Borzoi replicates are available at https://huggingface.co/johahi under the MIT license.

genomics↗

bioRxiv2024

Non-destructive transcriptomics via vesicular export

Transcriptomics enables comprehensive, multiplexed characterization of cellular states, yet prevailing methods typically require cell fixation or lysis, precluding longitudinal analysis of RNA expression in living cells. Here, we present non-destructive transcriptomics by vesicular export (NTVE), a platform for multi-time-point monitoring of RNA expression dynamics in living cells. Stabilized RNA reporter barcodes can be selectively packaged and exported from cells via virus-like particles (VLPs) bearing bioorthogonal affinity handles for convenient multichannel tracking of co-cultured cells. Using an engineered poly(A)-binding protein adapter, NTVE exports endogenous transcripts from inducible human and murine cell lines with high concordance to conventional lysate-derived RNA-seq. NTVE captures transcriptome changes in response to genetic and chemical perturbations within the same cells over time using standard sequencing workflows. NTVE can further be equipped with fusogens to deliver mRNA-encoded effectors or ribonucleoprotein gene editors from sender cells, activating gene reporters in co-cultured recipient cells. We demonstrate the utility of NTVE for monitoring hiPSC differentiation through daily non-destructive transcriptomic profiling of lineage-specific marker dynamics.

bioengineering↗

bioRxiv2024

scooby: Modeling multi-modal genomic profiles from DNA sequence at single-cell resolution

Understanding how regulatory DNA elements shape gene expression across individual cells is a fundamental challenge in genomics. Joint RNA-seq and epigenomic profiling provides opportunities to build unifying models of gene regulation capturing sequence determinants across steps of gene expression. However, current models, developed primarily for bulk omics data, fail to capture the cellular heterogeneity and dynamic processes revealed by single-cell multi-modal technologies. Here, we introduce scooby, the first framework to model scRNA-seq coverage and scATAC-seq insertion profiles along the genome from sequence at single-cell resolution. For this, we leverage the pre-trained multi-omics profile predictor Borzoi as a foundation model, equip it with a cell-specific decoder, and fine-tune its sequence embeddings. Specifically, we condition the decoder on the cell position in a precomputed single-cell embedding resulting in strong generalization capability. Applied to a hematopoiesis dataset, scooby recapitulates cell-specific expression levels of held-out genes, and identifies regulators and their putative target genes through in silico motif deletion. Moreover, accurate variant effect prediction with scooby allows for breaking down bulk eQTL effects into single-cell effects and delineating their impact on chromatin accessibility and gene expression. We anticipate scooby to aid unraveling the complexities of gene regulation at the resolution of individual cells.

genomics↗

bioRxiv2024

Nucleotide dependency analysis of DNA language models reveals genomic functional elements

Deciphering how nucleotides in genomes encode regulatory instructions and molecular machines is a long-standing goal in biology. DNA language models (LMs) implicitly capture functional elements and their organization from genomic sequences alone by modeling probabilities of each nucleotide given its sequence context. However, using DNA LMs for discovering functional genomic elements has been challenging due to the lack of interpretable methods. Here, we introduce nucleotide dependencies which quantify how nucleotide substitutions at one genomic position affect the probabilities of nucleotides at other positions. We generated genome-wide maps of pairwise nucleotide dependencies within kilobase ranges for animal, fungal, and bacterial species. We show that nucleotide dependencies indicate deleteriousness of human genetic variants more effectively than sequence alignment and DNA LM reconstruction. Regulatory elements appear as dense blocks in dependency maps, enabling the systematic identification of transcription factor binding sites as accurately as models trained on experimental binding data. Nucleotide dependencies also highlight bases in contact within RNA structures, including pseudoknots and tertiary structure contacts, with remarkable accuracy. This led to the discovery of four novel, experimentally validated RNA structures in Escherichia coli. Finally, using dependency maps, we reveal critical limitations of several DNA LM architectures and training sequence selection strategies by benchmarking and visual diagnosis. Altogether, nucleotide dependency analysis opens a new avenue for discovering and studying functional elements and their interactions in genomes.

genomics↗

bioRxiv2024

Functional gene embeddings improve rare variant polygenic risk scores

Rare variant association testing is a powerful strategy for identifying effector genes underlying common traits. However, its effectiveness is limited by the scarcity of high-impact rare allele carriers, posing challenges for sensitivity and robustness. Here, we introduce FuncRVP, a rare variant association framework addressing this issue by leveraging functional information across genes. FuncRVP models the effects of rare variants as a weighted sum of gene impairment scores, with weights regularized through a prior based on functional gene embeddings. Modeling 41 quantitative traits from unrelated UK Biobank participants showed that FuncRVP consistently outperformed linear regressions on significantly associated genes and did so more effectively for traits with higher burden heritability. The framework demonstrated versatility, yielding consistent improvements across diverse gene embeddings. Moreover, FuncRVP generated more robust gene effect estimates and yielded more gene discoveries, especially among genetically constrained genes. These findings demonstrate the value of integrating functional information in rare variant association studies and showcase FuncRVP as a promising tool for enhancing phenotype prediction and gene discovery.

genetics↗

bioRxiv2023

Aberrant expression prediction across human tissues

Despite the frequent implication of aberrant gene expression in diseases, algorithms predicting aberrantly expressed genes of an individual are lacking. To address this need, we compiled an aberrant expression prediction benchmark covering 8.2 million rare variants from 633 individuals across 49 tissues. While not geared toward aberrant expression, the deleteriousness score CADD and the loss-of-function predictor LOFTEE showed mild predictive ability (1-1.6% average precision). Leveraging these and further variant annotations, we next trained AbExp, a model that yielded 12% average precision by combining in a tissue-specific fashion expression variability with variant effects on isoforms and on aberrant splicing. Integrating expression measurements from clinically accessible tissues led to another two-fold improvement. Furthermore, we show on UK Biobank blood traits that performing rare variant association testing using the continuous and tissue-specific AbExp variant scores instead of LOFTEE variant burden increases gene discovery sensitivity and enables improved phenotype predictions.

genetics↗

bioRxiv2023

Identifying dysregulated regions in amyotrophic lateral sclerosis through DNA accessibility outliers

The high heritability of ALS and similar rare diseases contrasts with their low molecular diagnosis rate post-genetic testing, pointing to potential undiscovered genetic factors. Chromatin accessibility assays quantify the activity of functional elements genome-wide, offering invaluable insights into dysregulated regions. In this research, we introduced EpiOut, a computational toolbox to identify outliers in chromatin accessibility. These outliers represent dysregulated regions where chromatin accessibility uniquely diverges from the population baseline in a single or few samples. Annotation of accessible regions with histone ChIP-seq and Hi-C indicates that outliers are concentrated in functional loci, especially among promoters interacting with active enhancers. Across different omics levels, outliers are robustly replicated, and chromatin accessibility outliers are reliable predictors of gene expression outliers and aberrant protein levels. For example, 59% of gene expression outliers can be linked to aberration in chromatin accessibility. When promoter accessibility does not align with gene expression, our results indicate that molecular aberrations are more likely to be linked to post-transcriptional regulation rather than transcriptional regulation. Our findings demonstrate that the outlier detection paradigm can uncover dysregulated regions in rare diseases. EpiOut is open-sourced and freely available at github.com/uci-cbcl/EpiOut.

bioinformatics↗

bioRxiv2023

A comprehensive atlas of the aging vertebrate brain reveals signatures of progressive proteostasis dysfunction

Aging and neurodegeneration entail diverse cellular and molecular hallmarks. Here, we studied the effects of aging on the transcriptome, translatome, and multiple layers of the proteome in the brain of a short-lived killifish. We reveal that aging causes widespread reduction of proteins enriched in basic amino acids that is independent of mRNA regulation, and it is not due to impaired proteasome activity. Instead, we identify a cascade of events where aberrant translation pausing leads to reduced ribosome availability resulting in proteome remodeling independently of transcriptional regulation. Our research uncovers a vulnerable point in the aging brains biology - the biogenesis of basic DNA/RNA binding proteins. This vulnerability may represent a unifying principle that connects various aging hallmarks, encompassing genome integrity and the biosynthesis of macromolecules. One-Sentence SummaryTranslation pausing reshapes the aging brain proteome, revealing vulnerabilities in the biogenesis of nucleic-acid protein.

systems biology↗

bioRxiv2023

Integration of variant annotations using deep set networks boosts rare variant association genetics

Rare genetic variants can strongly predispose to disease, yet accounting for rare variants in genetic analyses is statistically challenging. While rich variant annotations hold the promise to enable well-powered rare variant association tests, methods integrating variant annotations in a data-driven manner are lacking. Here, we propose DeepRVAT, a model based on set neural networks that learns burden scores from rare variants, annotations, and phenotypes. In contrast to existing methods, DeepRVAT yields a single, trait-agnostic, nonlinear gene impairment score, enabling both risk prediction and gene discovery in a unified framework. On 34 quantitative and 26 binary traits, using whole-exome-sequencing data from UK Biobank, we find that DeepRVAT offers substantial increases in gene discoveries and improved replication rates in held-out data. Moreover, we demonstrate that the integrative DeepRVAT gene impairment score greatly improves detection of individuals at high genetic risk. Finally, we show that pre-trained DeepRVAT scores generalize across traits, opening up the possibility to conduct highly computationally efficient rare variant tests.

bioinformatics↗