Biology subjects

Berger, B.

Publications and source records attributed to Berger, B..

At least 19 recordsLinked to original sources

bioRxiv2026

BatchRefiner: fast, significant improvement in batch integration of single-cell embeddings with ensemble refinement

Data from single-cell RNA sequencing (scRNA-seq) and the Assay for Transposase-Accessible Chromatin (scATAC-seq) are high-dimensional, sparse, and undesirably capture technical variability between experiments or batches. Many analysis methods thus seek to produce a low-dimensional cell-by-feature embedding space that groups together biologically similar cells across batches while distancing dissimilar cells. Here, we introduce ensemble refinement for scRNA-seq and scATAC-seq embeddings, inspired by ensemble methods from statistical machine learning, and implement BatchRefiner, a fast post-processing tool to enhance batch integration. We extensively benchmark widely-used scRNA-seq embedding methods on both batch integration and biological conservation over a wide range of datasets, before and after the addition of BatchRefiner. We extend these benchmarking approaches to provide the first comprehensive benchmark of batch integration for scATAC-seq embedding methods, including BatchRefiner. Importantly, we formalize a significance statistic, which we use to demonstrate BatchRefiner's significant improvement in batch integration across a wide range of embedding methods, atlas-scale datasets, and established metrics.

bioinformatics↗

bioRxiv2026

Promera: a unified model for biomolecular structure prediction, filtering, and design

Generative models have become staple tools for modeling and designing biomolecular structures. However, although these tools have improved in structural prediction accuracy, their ability to filter designed binders--an essential use case--remains insufficient; whereas design methods have focused more on unconstrained binder generation rather than capabilities enabled by controllable design. We introduce Promera, a unified generative model that combines all-atom structure prediction with improved filtering and controllable design. We find that Promeras confidence metrics are more accurate for filtering binders from non-binders for both miniproteins and nanobodies, while its co-folding performance surpasses popular open-source models (OpenFold3-p2, Boltz-2) on therapeutically relevant categories. As a design model, Promera generates binders by predicting masked protein sequences with optional epitope, paratope, and template constraints. Remarkably, our nanobody designs match the in silico success rates from backprop-based techniques (mBER) when evaluated under co-folding confidence filters. We further provide two in silico demonstrations of the the versatile capabilities of our design method: epitope targeting of the Andes hantavirus glycoprotein with VHHs and active state stabilization of the {beta}2 andrenergic GPCR. We conclude by proposing a scaling law for co-folding models, suggesting a path for further performance improvement.

bioinformatics↗

bioRxiv2026

Thousandfold Expansion Microscopy

Biological macromolecules, such as proteins, are made of concatenated building blocks. We hypothesized that individual protein residues could be imaged by anchoring their side chains to a swellable polymer, cleaving backbone amide bonds, and expanding residues away from each other to a degree that enables them to be visualized separately. We introduce thousandfold expansion microscopy (1000ExM), a four-network interpenetrating hydrogel architecture that enables successive expansion from [~]18-fold to >1000-fold (one billion-fold in volume). Protein and peptide structures are maintained across these expansion factors, as verified by analyses of proteins with known structures (nanobodies, GFP) and a well-studied peptide (mCLING). Computational analysis indicates that 1000ExM resolves adjacent amino acid residues, thereby achieving sub-nanometer precision on conventional light microscopes. We anticipate that 1000ExM will find wide utility in protein visualization and identification, potentially even in intact cells and tissues.

molecular biology↗

bioRxiv2026

Evolutionary dynamics under phenotypic uncertainty

AO_SCPLOWBSTRACTC_SCPLOWClassical population genetics has largely relied on the same stochastic differential equations (SDEs) for over 60 years to describe evolutionary dynamics. However, these SDEs ignore the fact that phenotype heterogeneity and noise are ubiquitous in biological systems from bacteria to cancers. Here, we develop Probabilistic Phenotype Genetics (ProP Gen) theory as a mathematical framework for evolutionary dynamics under phenotypic uncertainty. Our newly introduced class of SDEs show that, remarkably, phenotypic uncertainty causes many central tenets of classical population genetics to break down, such as invariance of evolutionary dynamics to global shifts in absolute fitness at fixed population size. We revisit the long-studied problem of valley crossing in rugged fitness landscapes, discovering that lower-probability, high-fitness "phenotypic bridges" can substantially accelerate fitness valley crossing even at low mutation rates. We show our theory also explains a paradoxical concept we call "phenotypic buoying" whereby low-fitness phenotypes can exist at surprisingly high frequencies when carried by a high-fitness phenotype which acts as a source. ProP Gen theory uncovers complex phase diagrams of simultaneous coexistence between genotype-phenotype pairs due to phenotypic buoying, which we derive analytically exactly and verify numerically. Notably, the standard Wright-Fisher branching process is unsuitable for probabilistic phenotype numerics because it cannot correctly incorporate phenotypic uncertainty. Thus, we develop a more general, experimentally-inspired discrete-time simulation algorithm, Probabilistic Serial Dilution (ProSeD), which allows for overlapping generations, phenotypic noise, and stochastic phenotype switching. Finally, we show that the new diffusion limit of population genetics recapitulates the experimentally observed resuscitation and partitioning dynamics of bacterial "persister" strains. ProP Gen theory offers promise for describing and predicting the evolutionary dynamics of cancers, which have been empirically observed to exploit phenotypic uncertainty to evade treatment.

evolutionary biology↗

bioRxiv2026

Constrained Diffusion as a Paradigm for Evolution

A foundational question in computational biology is how to utilize data to describe the forces driving evolution. Here, we view evolution as a novel diffusion process constrained by many biological, physical, and environmental factors affecting organism viability at any given time. We introduce DiffEvol, a framework that models evolution as constrained diffusion over a discrete genotype space. Using real-world genomic sequence data alone, DiffEvol estimates complex evolutionary constraints by inverting the diffusion dynamics to recover a constrained subspace representing the viable genotype manifold, as well as its evolution over time. Applied to SARS-CoV-2 sequence data from 2020-2024, DiffEvol reconstructs constraint functions that recapitulate known viral fitness trends, including a pronounced "phase transition" that occurred following the widespread adoption of the SARS-CoV-2 vaccine. Our constraint subspace representation of the data characterizes such features and trends more clearly. This framework could be used not only to improve forecasting of emergent pathogenic strains, but also to produce more accurate reverse time analyses of their evolutionary dynamics to help identify ancestral variants and the forces having shaped a pathogens evolutionary trajectory. More generally, this formulation provides a method for linking observed sequence mutations to an evolving fitness landscape. Thus, constrained subspace diffusion offers a mathematical language for evolutionary dynamics in any system where random variation interacts with slowly-changing structural or global constraints, and can be applied to more complex evolutionary phenomena such as vaccine resistance, viral escape, and protein evolution.

bioinformatics↗

bioRxiv2025

Generating functional and multistate proteins with a multimodal diffusion transformer

Generating proteins with the full diversity and complexity of functions found in nature is a grand challenge in protein design. Here, we present ProDiT, a multimodal diffusion model that unifies sequence and structure modeling paradigms to enable the design of functional proteins at scale. Trained on sequences, 3D structures, and annotations for 214M proteins across the evolutionary landscape, ProDiT generates diverse, novel proteins that preserve known active and binding site motifs and can be successfully conditioned on a wide range of molecular functions, spanning 465 Gene Ontology terms. We introduce a diffusion sampling protocol to design proteins with multiple functional states, and demonstrate this protocol by scaffolding enzymatic active sites from carbonic anhydrase and lysozyme to be allosterically deactivated by a calcium effector. Our results showcase ProDiTs unique capacity to satisfy design specifications inaccessible to existing generative models, thereby expanding the protein design toolkit.

bioinformatics↗

bioRxiv2025

Tracing the Shared Foundations of Gene Expression and Chromatin Structure

The three-dimensional organization of chromatin into topologically associating domains (TADs) may impact gene regulation by bringing distant genes into contact. However, studies of TADs function and their influence on transcription have been constrained by ambiguities in TAD boundary definitions and challenges in directly measuring their regulatory effects. We overcome these limitations by developing species-level consensus TAD maps for human and mouse using a bag-of-genes approach that exposes emergent regulatory structure. To quantify TAD-mediated gene relationships, we leverage a generative AI foundation model computed from 33 million transcriptomes to define a contextual similarity metric, revealing higher-order gene relationships elusive to co-expression analysis. We find that TADs are regions of elevated gene co-regulation, and our analytical framework directly leads to testable hypotheses about chromatin organization across cellular states. We discover that the TAD-linked enhancement of transcriptional context is strongest in early developmental stages and systematically declines with aging. Investigation of cancer cells show distinct patterns of TAD usage that shift with chemotherapy treatment. Enhancing our understanding of cellular plasticity in differentiation and disease, these findings suggest that chromatin organization may act through probabilistic mechanisms rather than deterministic rules. Software availabilityhttps://singhlab.net/tadmap

bioinformatics↗

bioRxiv2025

Learning the language of protein-protein interactions

Protein Language Models (PLMs) trained on large databases of protein sequences have proven effective in modeling protein biology across a wide range of applications. However, while PLMs excel at capturing individual protein properties, they face challenges in natively representing protein-protein interactions (PPIs), which are crucial to understanding cellular processes and disease mechanisms. Here, we introduce MINT, a PLM specifically designed to model sets of interacting proteins in a contextual and scalable manner. Using unsupervised training on a large curated PPI dataset derived from the STRING database, MINT outperforms existing PLMs in diverse tasks relating to protein-protein interactions, including binding affinity prediction and estimation of mutational effects. Beyond these core capabilities, it excels at modeling interactions in complex protein assemblies and surpasses specialized models in antibody-antigen modeling and T cell receptor-epitope binding prediction. MINTs predictions of mutational impacts on oncogenic PPIs align with experimental studies, and it provides reliable estimates for the potential for cross-neutralization of antibodies against SARS-CoV-2 variants of concern. These findings position MINT as a powerful tool for elucidating complex protein interactions, with significant implications for biomedical research and therapeutic discovery.

bioinformatics↗

bioRxiv2025

ralphi: a deep reinforcement learning framework for haplotype assembly

Haplotype assembly is the problem of reconstructing the combination of alleles on the maternally and paternally inherited chromosome copies. Individual haplotypes are essential to our understanding of how combinations of different variants impact phenotype. In this work, we focus on read-based haplotype assembly of individual diploid genomes, which reconstructs the two haplotypes directly from read alignments at variant loci. We introduce ralphi, a novel deep reinforcement learning framework for haplotype assembly, which integrates the representational power of deep learning with reinforcement learning to accurately partition read fragments into their respective haplotype sets. To set the reward objective for reinforcement learning, our approach uses the classic reduction of the problem to the maximum fragment cut formulation on fragment graphs, where nodes correspond to reads and edge weights capture the conflict or agreement of the reads at shared variant sites. We trained ralphi on a diverse dataset of fragment graph topologies derived from genomes in the 1000 Genomes Project. We show that ralphi consistently achieves lower error rates at comparable or longer haplotype block lengths over the state of the art for short and long ONT reads at varying coverage in standard human genome benchmarks. ralphi is available at https://github.com/PopicLab/ralphi.

bioinformatics↗

bioRxiv2024

Decoding the Functional Interactome of Non-Model Organisms with PHILHARMONIC

Despite the widespread availability of genome sequencing pipelines, many genes remain part of the genome's "dark matter," where existing inference tools cannot even begin to guess the biological function of their proteins from sequence alone. This challenge is especially pronounced in organisms that are highly evolutionarily distant from well-studied models, where homology-based methods break down. Here, we describe PHILHARMONIC, a computational method that combines deep learning-based de novo protein interaction network inference with robust unsupervised spectral clustering and remote homology to illuminate functional organization in any non-model organism. From only a sequenced proteome, we show PHILHARMONIC generates meaningful functional hypothesis for protein functions, functional communities, and higher-order network structure. We validate its performance using experimental gene expression and pathway data in D. melanogaster, and we demonstrate its broad utility by analyzing temperature sensing and stress response pathways in the reef-building coral P. damicornis and its algal symbiont C. goreaui. PHILHARMONIC provides a general-purpose engine for functional discovery and biological hypothesis generation in non-model organisms, enabling systems-level insights across the full diversity of life.

bioinformatics↗

bioRxiv2024

Effect of Mutations on Smlt1473 Binding to Various Substrates Using Molecular Dynamics Simulations

Smlt1473 is a polysaccharide lyase from Stenotrophomonas maltophilia whose crystal structure was solved recently using X-ray crystallography. There was an effort to study the effect of mutations on the activity of Smlt1473 binding to various substrates like hyaluronic acid(HA), mannuronic acid(ManA), and alginate. In this study, we use molecular docking and molecular dynamics simulations to investigate the effect of binding of various substrates (HA and ManA) to Smlt1473 and two of its mutants H221F and R312L. We further studied the stability in the binding of Smlt1473 to its various substrates as well as the role of fluctuations. Machine-Learning based clustering algorithms were used to group the entire simulation trajectory into various stable states. The molecular interactions Smlt1473 to the substrates were calculated and the importance of specific residues were tested with observed activity assays due to residue mutations. Overall, we find that the R218 plays an important role in substrate binding and thus impacting the activity due to the H221F mutant and R/L312 itself plays an important role in the R312 mutation. In addition, we have also found three more residues K56, R107, and R164 important for substrate binding which we further proceed to confirm using wet lab mutagenesis studies.

biophysics↗

bioRxiv2024

Secure Discovery of Genetic Relatives across Large-Scale and Distributed Genomic Datasets

Finding relatives within a study cohort is a necessary step in many genomic studies. However, when the cohort is distributed across multiple entities subject to data-sharing restrictions, performing this step often becomes infeasible. Developing a privacy-preserving solution for this task is challenging due to the significant burden of estimating kinship between all pairs of individuals across datasets. We introduce SF-Relate, a practical and secure federated algorithm for identifying genetic relatives across data silos. SF-Relate vastly reduces the number of individual pairs to compare while maintaining accurate detection through a novel locality-sensitive hashing approach. We assign individuals who are likely to be related together into buckets and then test relationships only between individuals in matching buckets across parties. To this end, we construct an effective hash function that captures identity-by-descent (IBD) segments in genetic sequences, which, along with a new bucketing strategy, enable accurate and practical private relative detection. To guarantee privacy, we introduce an efficient algorithm based on multiparty homomorphic encryption (MHE) to allow data holders to cooperatively compute the relatedness coefficients between individuals, and to further classify their degrees of relatedness, all without sharing any private data. We demonstrate the accuracy and practical runtimes of SF-Relate on the UK Biobank and All of Us datasets. On a dataset of 200K individuals split between two parties, SF-Relate detects 94.9% of third-degree relatives, and 99.9% of second-degree or closer relatives, within 15 hours of runtime. Our work enables secure identification of relatives across large-scale genomic datasets.

bioinformatics↗

bioRxiv2023

Democratizing Protein Language Models with Parameter-Efficient Fine-Tuning

Proteomics has been revolutionized by large pre-trained protein language models, which learn unsupervised representations from large corpora of sequences. The parameters of these models are then fine-tuned in a supervised setting to tailor the model to a specific downstream task. However, as model size increases, the computational and memory footprint of fine-tuning becomes a barrier for many research groups. In the field of natural language processing, which has seen a similar explosion in the size of models, these challenges have been addressed by methods for parameter-efficient fine-tuning (PEFT). In this work, we newly bring parameter-efficient fine-tuning methods to proteomics. Using the parameter-efficient method LoRA, we train new models for two important proteomic tasks: predicting protein-protein interactions (PPI) and predicting the symmetry of homooligomers. We show that for homooligomer symmetry prediction, these approaches achieve performance competitive with traditional fine-tuning while requiring reduced memory and using three orders of magnitude fewer parameters. On the PPI prediction task, we surprisingly find that PEFT models actually outperform traditional fine-tuning while using two orders of magnitude fewer parameters. Here, we go even further to show that freezing the parameters of the language model and training only a classification head also outperforms fine-tuning, using five orders of magnitude fewer parameters, and that both of these models outperform state-of-the-art PPI prediction methods with substantially reduced compute. We also demonstrate that PEFT is robust to variations in training hyper-parameters, and elucidate where best practices for PEFT in proteomics differ from in natural language processing. Thus, we provide a blueprint to democratize the power of protein language model tuning to groups which have limited computational resources.

bioinformatics↗

bioRxiv2023

Deciphering Salt Stress Responses in Solanum pimpinellifolium through High-Throughput Phenotyping.

Soil salinity is a major environmental stressor affecting agricultural productivity worldwide. Understanding plant responses to salt stress is crucial for developing resilient crop varieties. Wild relatives of cultivated crops, such as wild tomato, Solanum pimpinellifolium, can serve as a useful resource to further expand the resilience potential of the cultivated germplasm, S. lycopersicum. In this study, we employed high-throughput phenotyping in the greenhouse and field conditions to explore salt stress responses of a S. pimpinellifolium diversity panel. Our study revealed extensive phenotypic variations in response to salt stress, with traits such as transpiration rate, shoot mass, and ion accumulation showing significant correlations with plant performance. We found that while transpiration was a key determinant of plant performance in the greenhouse, shoot mass strongly correlated with yield under field conditions. Conversely, ion accumulation was the least influential factor under greenhouse conditions. Through a Genome Wide Association Study, we identified candidate genes not previously associated with salt stress, highlighting the power of high-throughput phenotyping in uncovering novel aspects of plant stress responses. This study contributes to our understanding of salt stress tolerance in S. pimpinellifolium and lays the groundwork for further investigations into the genetic basis of these traits, ultimately informing breeding efforts for salinity tolerance in tomato and other crops.

plant biology↗

bioRxiv2023

Learning the Language of Antibody Hypervariability

Protein language models (PLMs) based on machine learning have demon-strated impressive success in predicting protein structure and function. However, general-purpose ("foundational") PLMs have limited performance in predicting antibodies due to the latters hypervariable regions, which do not conform to the evolutionary conservation principles that such models rely on. In this study, we propose a new transfer learning framework called AbMAP, which fine-tunes foundational models for antibody-sequence inputs by supervising on antibody structure and binding specificity examples. Our feature representations accurately predict an antibodys 3D structure, mutational effects on antigen binding, and paratope identification. AbMAPs scalability paves the way for large-scale analyses of human antibody repertoires. AbMAP representations of immune repertoires reveal a remarkable overlap across individuals, overcoming the limitations of sequence analyses. Our findings provide compelling evidence for the hypothesis that antibody repertoires of individuals tend to converge towards comparable structural and functional coverage. We validate AbMAP for antibody optimization, applying it to optimize a set of antibodies that bind to a SARS-CoV-2 peptide and obtaining 82% hit-rate and upto 22-fold increase in binding affinity. We anticipate AbMAP will accelerate the efficient design and modeling of antibodies and expedite the discovery of antibody-based therapeutics. Availability:https://github.com/rs239/ablm

bioinformatics↗

bioRxiv2023

Assessing transcriptomic re-identification risks using discriminative sequence models

Gene expression data provides molecular insights into the functional impact of genetic variation, for example through expression quantitative trait loci (eQTL). With an improving understanding of the association between genotypes and gene expression comes a greater concern that gene expression profiles could be matched to genotype profiles of the same individuals in another dataset, known as a linking attack. Prior works demonstrating such a risk could analyze only a fraction of eQTLs that are independent due to restrictive model assumptions, leaving the full extent of this risk incompletely understood. To address this challenge, we introduce the discriminative sequence model (DSM), a novel probabilistic framework for predicting a sequence of genotypes based on gene expression data. By modeling the joint distribution over all known eQTLs in a genomic region, DSM improves the power of linking attacks with necessary calibration for linkage disequilibrium and redundant predictive signals. We demonstrate greater linking accuracy of DSM compared to existing approaches across a range of attack scenarios and datasets including up to 22K individuals, suggesting that DSM helps uncover a substantial additional risk overlooked by previous studies. Our work provides a unified framework for assessing the privacy risks of sharing diverse omics datasets beyond transcriptomics.

genomics↗

bioRxiv2023

Split-intein Gal4 provides intersectional genetic labeling that is fully repressible by Gal80

The split-Gal4 system allows for intersectional genetic labeling of highly specific cell-types and tissues in Drosophila. However, the existing split-Gal4 system, unlike the standard Gal4 system, cannot be repressed by Gal80, and therefore cannot be controlled temporally. This lack of temporal control precludes split-Gal4 experiments in which a genetic manipulation must be restricted to specific timepoints. Here, we describe a new split-Gal4 system based on a self-excising split-intein, which drives transgene expression as strongly as the current split-Gal4 system and Gal4 reagents, yet which is fully repressible by Gal80. We demonstrate the potent inducibility of "split-intein Gal4" in vivo using both fluorescent reporters and via reversible tumor induction in the gut. Further, we show that our split-intein Gal4 can be extended to the drug-inducible GeneSwitch system, providing an independent method for intersectional labeling with inducible control. We also show that the split-intein Gal4 system can be used to generate highly cell-type specific genetic drivers based on in silico predictions generated by single cell RNAseq (scRNAseq) datasets, and we describe a new algorithm ("Two Against Background" or TAB) to predict cluster-specific gene pairs across multiple tissue-specific scRNA datasets. We provide a plasmid toolkit to efficiently create split-intein Gal4 drivers based on either CRISPR knock-ins to target genes or using enhancer fragments. Altogether, the split-intein Gal4 system allows for the creation of highly specific intersectional genetic drivers that are inducible/repressible. Significance statementThe split-Gal4 system allows Drosophila researchers to drive transgene expression with extraordinary cell type specificity. However, the existing split-Gal4 system cannot be controlled temporally, and therefore cannot be applied to many important areas of research. Here, we present a new split-Gal4 system based on a self-excising split-intein, which is fully controllable by Gal80, as well as a related drug-inducible split GeneSwitch system. This approach can both leverage and inform single-cell RNAseq datasets, and we introduce an algorithm to identify pairs of genes that precisely and narrowly mark a desired cell cluster. Our split-intein Gal4 system will be of value to the Drosophila research community, and allow for the creation of highly specific genetic drivers that are also inducible/repressible.

genetics↗

bioRxiv2023

An econometric lens resolves cell-state parallax

Genome-wide association studies (GWAS) identify numerous disease-linked genetic variants at noncoding genomic loci, yet therapeutic progress is hampered by the challenge of deciphering the regulatory roles of these loci in tissue-specific contexts. Single-cell multimodal assays that simultaneously profile chromatin accessibility and gene expression could predict tissue-specific causal links between noncoding loci and the genes they affect. However, current computational strategies either neglect the causal relationship between chromatin accessibility and transcription or lack variant-level precision, aggregating data across genomic ranges due to data sparsity. To address this, we introduce GrID-Net, a graph neural network approach that generalizes Granger causal inference to detect new causal locus-gene associations in graph-structured systems such as single-cell trajectories. Inspired by the principles of optical parallax, which reveals object depth from static snapshots, we hypothesized that causal mechanisms could be inferred from static single-cell snapshots by exploiting the time lag between epigenetic and transcriptional cell states, a concept we term "cell-state parallax." Applying GrID-Net to schizophrenia (SCZ) genetic variants, we increased variant coverage by 36% and uncovered noncoding mechanisms that dysregulate 132 genes, including key potassium transporters such as KCNG2 and SLC12A6. Furthermore, we discovered evidence for the prominent role of neural transcription-factor binding disruptions in SCZ etiology. Our work not only provides a strategy for elucidating the tissue-specific impact of noncoding variants but also underscores the breakthrough potential of cell-state parallax in single-cell multiomics for discovering tissue-specific gene regulatory mechanisms.

bioinformatics↗