Search bioRxiv⌕ Search

Biology subjects

Song, Y. S.

Publications and source records attributed to Song, Y. S..

At least 19 recordsLinked to original sources
bioRxiv2026

NovaClone: A Network-Based Algorithm for Clonal and Subclonal Genotyping of Barcoded Transgene Integrations

Single-cell lineage tracing technologies are providing new and powerful ways to interrogate the evolution and divergence of cell populations in cancer, development, and other contexts. A key initial step in any such analysis is the grouping of cells into clonal populations, based on clone-level marks. Unfortunately, clone calling is prone to technical effects due to sequencing errors, missing data, multiplets, background noise, and accidental sharing of clonal barcodes between unrelated clones (homoplasy). We present NovaClone, a principled algorithm for hierarchical clone calling that is broadly applicable to all current tracing technologies, including both static barcoding and the more recent evolving tracers. We benchmark NovaClone on simulated and real data to show that it outperforms the current solutions in terms of both quality and speed, thereby helping to mitigate one of the most prevalent problems with single-cell lineage tracing. To complement NovaClone, we introduce a suite of algorithm-agnostic quality control metrics to evaluate clone calls when ground truth is not available. NovaClone and the associated QCs are available through the open source Python package nova-clone.

genomics↗
bioRxiv2026

Deep models of protein evolution in time generate realistic evolutionary trajectories and functional proteins

Models of protein evolution are foundational to biology, underpinning essential techniques such as phylogenetic tree inference, ancestral sequence reconstruction, multiple sequence alignment, variant effect prediction, and protein design. Historically, for computational tractability, these models have relied on the simplifying - but biologically unrealistic - assumption that sites in a given protein evolve independently of each other. A crucial test of any evolutionary model is its ability to simulate realistic evolutionary trajectories, but the independent-sites assumption leads to simulations that poorly reflect the complexity of natural protein evolution. Here we introduce PEINT (Protein Evolution IN Time), a flexible and generalizable deep learning framework for modeling how the entire protein sequence evolves over time while incorporating complex interactions between sites. This framework enables learning realistic patterns of constrained evolutionary transitions directly from millions of protein sequences spanning diverse fold families. Furthermore, unlike classical models that require pre-aligned sequences, PEINT learns indel dynamics directly from raw, unaligned sequences, thereby eliminating potential biases from alignment errors that can lead to incorrect inference of evolutionary patterns. By capturing higher-order epistatic interactions and modeling insertion-deletion processes that classical models typically ignore, PEINT accurately reproduces key signatures of natural evolution, including conservation patterns and family-specific dynamics. When simulating evolution along phylogenetic trees, PEINT generates highly novel sequences that preserve protein function, which we validate through experimental characterization of simulated carbonic anhydrase variants that retain enzymatic activity. PEINT thus enables realistic simulation of protein evolution that explores new sequence space while respecting structural and functional constraints. This evolution-informed generative modeling framework offers a powerful new tool for advancing both phylogenetic inference and protein engineering.

evolutionary biology↗
bioRxiv2026

Architecting cis-regulation to quantitatively tune gene expression in cereals

Precise modulation of gene expression via cis-regulatory editing holds promise for non-transgenic crop improvement, but the sequence-to-function relationships that govern plant promoter activity remain poorly understood. Here, we develop a massively parallel reporter assay (MPRA) in Sorghum bicolor to systematically measure the effects of >30,000 CRISPR-like mutations-deletions, substitutions, and motif insertions-across entire native promoters and 5' untranslated regions (UTRs) of three photosynthesis genes: PsbS, Raf1, and SBPase. We find that gene expression is most tunable within a [~]500 base pair core promoter region, where mutational effects are reproducible across biological replicates and predictive of protein output. Within these regions, we identify compact deletions and motif insertions that strongly increase protein production (>30-fold relative to wild type), exceeding the performance of transgenic enhancer elements. Mutation-effect relationships are gene-specific, highlighting the need for tailored regulatory maps. Our results establish a high-throughput strategy for cis-regulatory fine-mapping that enables crop improvements via minimal, precise, and non-transgenic gene edits.

synthetic biology↗
bioRxiv2025

Enhanced liver regeneration via targeted mRNA delivery for partial in vivo reprogramming

Recent studies suggest that injury-induced dedifferentiation, which leads to the formation of injury-responsive cells, contributes significantly to tissue repair across various organs, including the liver. Utilizing Yamanaka factors (Oct4, Sox2, Klf4, and c-Myc: OSKM) for in vivo partial reprogramming generates injury-responsive cells in the intestine, mirroring those derived from injury-induced dedifferentiation. Thus, the transgene induction of OSKM or viral delivery of Oct4, Sox2, and Klf4 shows promise in facilitating tissue regeneration in the intestine, liver, skeletal muscle, and retina. Herein, we demonstrated that transient OSKM induction produces two distinct liver progenitor-like cell populations. One of these populations resembles liver progenitor-like cells (LPLCs) generated by acute acetaminophen (APAP) injury without triggering immune responses. To explore in vivo reprogramming as a viable strategy for tissue regeneration, we employed lipid nanoparticles (LNP) carrying OSKM mRNA (OSKM mRNA-LNP) to stimulate LPLCs formation. Notably, the production of Sox9+ LPLCs, and OSKM-induced dedifferentiation, was closely correlated with successful tissue regeneration in the liver post APAP injury. Thus, the OSKM mRNA-LNP approach represents a promising therapeutic intervention for the repair of acute liver injuries.

physiology↗
bioRxiv2025

Separating selection from mutation in antibody language models

Antibodies are encoded by nucleotide sequences that are generated by V(D)J recombination and evolve according to mutation and selection processes. Existing antibody language models, however, focus exclusively on antibodies as strings of amino acids and are fitted using standard language modeling objectives such as masked or autoregressive prediction. In this paper, we first show that fitting models using this objective implicitly incorporates nucleotide-level mutation processes as part of the protein language model, which degrades performance when predicting effects of mutations on functional properties of antibodies. To address this limitation, we devise a new framework: a Deep Amino acid Selection Model (DASM) that learns the selection effects of amino-acid mutations while explicitly factoring out the nucleotide-level mutation process. By fitting selection as a separate term from the mutation process, the DASM exclusively quantifies functional effects: effects that change some aspect of the function of the antibody. This factorization leads to substantially improved performance on standard functional benchmarks. Moreover, our model is an order of magnitude smaller and multiple orders of magnitude faster to evaluate than existing approaches, as well as being readily interpretable.

evolutionary biology↗
bioRxiv2025

Predicting functional constraints across evolutionary timescales with phylogeny-informed genomic language models

Genomic language models (gLMs) have emerged as a powerful approach for learning genome-wide functional constraints directly from DNA sequences. However, standard gLMs adapted from natural language processing often require extremely large model sizes and computational resources, yet still fall short of classical evolutionary models in predictive tasks. Here, we introduce GPN-Star (Genomic Pretrained Network with Species Tree and Alignment Representation), a biologically grounded gLM featuring a phylogeny-aware architecture that leverages whole-genome alignments and species trees to model evolutionary relationships explicitly. Trained on alignments spanning vertebrate, mammalian, and primate evolutionary timescales, GPN-Star achieves state-of-the-art performance across a wide range of variant effect prediction tasks in both coding and non-coding regions of the human genome. Analyses across timescales reveal task-dependent advantages of modeling more recent versus deeper evolution. To demonstrate its potential to advance human genetics, we show that GPN-Star substantially outperforms prior methods in prioritizing pathogenic and fine-mapped GWAS variants; yields unprecedented enrichments of complex trait heritability; and improves power in rare variant association testing. Extending beyond humans, we train GPN-Star for five model organisms - Mus musculus, Gallus gallus, Drosophila melanogaster, Caenorhabditis elegans, and Arabidopsis thaliana - demonstrating the robustness and generalizability of the framework. Taken together, these results position GPN-Star as a scalable, powerful, and flexible new tool for genome interpretation, well suited to leverage the growing abundance of comparative genomics data.

bioinformatics↗
bioRxiv2025

Inference of germinal center evolutionary dynamics via simulation-based deep learning

B cells and the antibodies they produce are vital to health and survival, motivating research on the details of the mutational and evolutionary processes in the germinal centers (GCs) from which mature B cells arise. It is known that B cells with higher affinity for their cognate antigen (Ag) will, on average, tend to have more offspring. However, the exact form of this relationship between affinity and fecundity, which we call the "affinity-fitness response function", is not known. Here we use deep learning and simulation-based inference to learn this function from a unique experiment that replays a particular combination of GC conditions many times. All code is freely available at https://github.com/matsengrp/gcdyn, while datasets and inference results can be found at https://doi.org/10.5281/zenodo.15022130.

evolutionary biology↗
bioRxiv2025

Nucleotide context models outperform protein language models for predicting antibody affinity maturation

Antibodies play a crucial role in adaptive immunity. They develop as B cell receptors (BCRs): membrane-bound forms of antibodies that are expressed on the surfaces of B cells. BCRs are refined through affinity maturation, a process of somatic hypermutation (SHM) and natural selection, to improve binding to an antigen. Computational models of affinity maturation have developed from two main perspectives: molecular evolution and language modeling. The molecular evolution perspective focuses on nucleotide sequence context to describe mutation and selection; the language modeling perspective involves learning patterns from large data sets of protein sequences. In this paper, we compared models from both perspectives on their ability to predict the course of antibody affinity maturation along phylogenetic trees of BCR sequences. This included models of SHM, models of SHM combined with an estimate of selection, and protein language models. We evaluated these models for large human BCR repertoire data sets, as well as an antigen-specific mouse experiment with a pre-rearranged cognate naive antibody. We demonstrated that precise modeling of SHM, which requires the nucleotide context, provides a substantial amount of predictive power for predicting the course of affinity maturation. Notably, a simple nucleotide-based convolutional neural network modeling SHM outperformed state-of-the-art protein language models, including one trained exclusively on antibody sequences. Furthermore, incorporating estimates of selection based on a custom deep mutational scanning experiment brought only modest improvement in predictive power. To support further research, we introduce EPAM (Evaluating Predictions of Affinity Maturation), a benchmarking framework to integrate evolutionary principles with advances in language modeling, offering a road map for understanding antibody evolution and improving predictive models.

immunology↗
bioRxiv2025

Replaying germinal center evolution on a quantified affinity landscape

Darwinian evolution of immunoglobulin genes within germinal centers (GCs) underlies the progressive increase in antibody affinity following antigen exposure. Whereas the cellular mechanics of how competition between B cells produces increases in affinity are well established, the evolutionary dynamics of this process are less clear. We developed an experimental evolution model where we "replay" over one hundred monoclonal GC reactions, assigning affinities to each cell using deep mutational scanning. Our data reveal how GCs achieve predictable outcomes by means of noisy but persistent selection on an affinity landscape whose exploration is heavily constrained by somatic hypermutation biases. We infer a fitness landscape that quantitatively recapitulates the affinity maturation trajectory of our clone and find that apparent features of GC selection such as permissiveness to low-affinity lineages and rapid plateauing of affinity are likely artifacts of survivorship biases that distort our view of how B cell affinity progresses over time.

immunology↗
bioRxiv2025

Robust and Adaptive Non-Parametric Tests for Detecting General Distributional Shifts in Gene Expression

Differential expression analysis is crucial in genomics, yet existing methods primarily focus on detecting mean shifts. Variance shifts in gene expression are well-documented in studies of cellular signaling pathways, and more recently they have characterized aging, thus motivating the need for flexible detection approaches that include tests of expression variance changes. In this work, we present QRscore (Quantile Rank Score), a general method for detecting distributional shifts in gene expression by extending the Mann-Whitney test into a flexible family of rank-based tests. Here, we focus on implementing QRscore to detect shifts in mean and variance in gene expression, using weights designed from negative binomial (NB) and zero-inflated negative binomial (ZINB) models to combine the strengths of parametric and non-parametric approaches. We show through simulations that QRscore not only achieves high statistical power while controlling the false discovery rate (FDR), but also outperforms existing methods in detecting variance shifts and mean shifts. Applying QRscore to bulk RNA-seq data from the Genotype-Tissue Expression (GTEx) project, we identified numerous differentially dispersed genes and differentially expressed genes across 33 tissues. Notably, many genes have significant variance shifts but non-significant mean shifts. QRscore augments the genome bioinformatics toolkit by offering a powerful and flexible approach for differential expression analysis. QRscore is available in R, at https://github.com/songlab-cal/QRscore.

bioinformatics↗
bioRxiv2025

A General Framework for Branch Length Estimation in Ancestral Recombination Graphs

Inference of Ancestral Recombination Graphs (ARGs) is of central interest in the analysis of genomic variation. ARGs can be specified in terms of topologies and coalescence times. The coalescence times are usually estimated using an informative prior derived from coalescent theory, but this may generate biased estimates and can also complicate downstream inferences based on ARGs. Here we introduce, POLEGON, a novel approach for estimating branch lengths for ARGs which uses an uninformative prior. Using extensive simulations, we show that this method provides improved estimates of coalescence times and lead to more accurate inferences of effective population sizes under a wide range of demographic assumptions (population expansion, bottleneck, split, etc). It also improves other downstream inferences including estimates of mutation rates. We apply the method to data from the 1000 Genomes Project to investigate population size histories and differential mutation signatures across populations. We also estimate coalescence times in the HLA region, and show that they exceed 30 million years in multiple segments. Significance StatementModel misspecification is a common challenge in population genetic inference, as oversimplified mathematical models often fail to capture complex evolutionary processes. Here we introduce a novel framework for branch length estimation in whole-genome genealogies, which enables accurate inference using non-informative priors and subsequent posterior calibration, rather than relying on heavily parametrized models. This flexible approach has been validated in various simulation settings. When applied to real genomic data, it enables robust inference of demography, mutation rates, and transspecies polymorphisms.

evolutionary biology↗
bioRxiv2025

Benchmarking DNA Sequence Models for Causal Regulatory Variant Prediction in Human Genetics

Machine learning holds immense promise in biology, particularly for the challenging task of identifying causal variants for Mendelian and complex traits. Two primary approaches have emerged for this task: supervised sequence-to-function models trained on functional genomics experimental data and self-supervised DNA language models that learn evolutionary constraints on sequences. However, the field currently lacks consistently curated datasets with accurate labels, especially for non-coding variants, that are necessary to comprehensively benchmark these models and advance the field. In this work, we present TraitGym, a curated dataset of regulatory genetic variants that are either known to be causal or are strong candidates across 113 Mendelian and 83 complex traits, along with carefully constructed control variants. We frame the causal variant prediction task as a binary classification problem and benchmark various models, including functional-genomics-supervised models, self-supervised models, models that combine machine learning predictions with curated annotation features, and ensembles of these. Our results provide insights into the capabilities and limitations of different approaches for predicting the functional consequences of non-coding genetic variants. We find that alignment-based models CADD and GPN-MSA compare favorably for Mendelian traits and complex disease traits, while functional-genomics-supervised models Enformer and Borzoi perform better for complex non-disease traits. Evo2 shows substantial performance gains with scale, but still lags somewhat behind alignment-based models, struggling particularly with enhancer variants. The benchmark, including a Google Colab notebook to evaluate a model in a few minutes, is available at https://huggingface.co/datasets/songlab/TraitGym.

genetics↗
bioRxiv2025

CPI-Pred: A deep learning framework for predicting functional parameters of compound-protein interactions

Recent advancements in deep learning have enabled functional annotation of genome sequences, facilitating the discovery of new enzymes and metabolites. However, accurately predicting compound-protein interactions (CPI) from sequences remains challenging due to the complexity of these interactions and the sparsity and heterogeneity of available data, which constrain the generalization of patterns across their solution space. In this work, we introduce CPI-Pred, a versatile deep learning model designed to predict compound-protein interaction function. CPI-Pred integrates compound representations derived from a novel message-passing neural network and enzyme representations generated by state-of-the-art protein language models, leveraging innovative sequence pooling and cross-attention mechanisms. To train and evaluate CPI-Pred, we compiled the largest dataset of enzyme kinetic parameters to date, encompassing four key metrics: the Michaelis-Menten constant (KM), enzyme turnover number (kcat), catalytic efficiency (kcat/KM), and inhibition constant (KI).These kinetic parameters are critical for elucidating enzyme function in metabolic contexts and understanding their regulation by compounds within biological networks. We demonstrate that CPI-Pred can predict diverse types of CPI using only the amino acid sequence of enzymes and structural representations of compounds, outperforming state-of-the-art models on unseen compounds and structurally dissimilar enzymes. Over workflow provides a valuable tool for tackling a range of metabolic engineering challenges, including the designing of novel enzyme sequences and compounds, such as enzyme inhibitors. Additionally, the datasets curated in this study offer a valuable resource for the scientific community, serving as a benchmark for machine learning models focused on enzyme activity and promiscuity prediction.

bioinformatics↗
bioRxiv2024

Learning antibody sequence constraints from allelic inclusion

Antibodies and B-cell receptors (BCRs) are produced by B cells, and are built of a heavy chain and a light chain. Although each B cell could express two different heavy chains and four different light chains, usually only a unique pair of heavy chain and light chain is expressed--a phenomenon known as allelic exclusion. However, a small fraction of naive-B cells violate allelic exclusion by expressing two productive light chains, one of which has impaired function; this has been called allelic inclusion. We demonstrate that these B cells can be used to learn constraints on antibody sequence. Using large-scale single-cell sequencing data from humans, we find examples of light chain allelic inclusion in thousands of naive-B cells, which is an order of magnitude larger than existing datasets. We train machine learning models to identify the abnormal sequences in these cells. The resulting models correlate with antibody properties that they were not trained on, including polyreactivity, surface expression, and mutation usage in affinity maturation. These correlations are larger than what is achieved by existing antibody modeling approaches, indicating that allelic inclusion data contains useful new information. We also investigate the impact of similar selection forces on the heavy chain in mouse, and observe that pairing with the surrogate light chain significantly restricts heavy chain diversity.

immunology↗
bioRxiv2024

Tree reconstruction guarantees from CRISPR-Cas9 lineage tracing data using Neighbor-Joining

CRISPR-Cas9 based lineage tracing technologies have enabled the reconstruction of single-cell phylogenies from transcriptional readouts. However, developing tree-reconstruction algorithms with theoretical guarantees in this setting is challenging. In this work, we derive a reconstruction algorithm with theoretical guarantees using Neighbor-Joining (NJ) on distances that are moment-matched to estimate the true tree distances. We develop a series of tools to analyze this algorithm and prove its theoretical guarantees. When the parameters of the data generating process are known and there is no missing data, our results align with established results from common evolutionary models, such as Cavender-Farris-Neyman and Jukes-Cantor. However, to account for the realistic case where the parameters of the data generating process are not known and there is missing data, we develop new theory that shows for the first time that it is still possible to obtain reconstruction guarantees in the CRISPR-Cas9 case and in other models of evolution. Empirically, we show on both simulated lineage tracing data and on real data from a mouse model of lung cancer the improved performance of our method as compared to the traditional use of NJ.

bioinformatics↗
bioRxiv2024

Dissecting the oncogenic mechanisms of POT1 cancer mutations through deep scanning mutagenesis

Mutations in the shelterin protein POT1 are associated with diverse cancers, but their role in cancer progression remains unclear. To resolve this, we performed deep scanning mutagenesis in POT1 locally haploid human stem cells to assess the impact of POT1 variants on cellular viability and cancer-associated telomeric phenotypes. Though POT1 is essential, frame-shift mutants are rescued by chemical ATR inhibition, indicating that POT1 is not required for telomere replication or lagging strand synthesis. In contrast, a substantial fraction of clinically-validated pathogenic mutations support normal cellular proliferation, but still drive ATR-dependent telomeric DNA damage signaling and ATR-independent telomere elongation. Moreover, this class of cancer-associated POT1 variants elongates telomeres more rapidly than POT1 frame-shifts, indicating they actively drive oncogenesis and are not simple loss-of-function mutations.

cancer biology↗
bioRxiv2024

Robust and Accurate Bayesian Inference of Genome-Wide Genealogies for Large Samples

The Ancestral Recombination Graph (ARG), which describes the full genealogical history of a sample of genomes, is a vital tool in population genomics and biomedical research. Recent advancements have increased ARG reconstruction scalability to tens or hundreds of thousands of genomes, but these methods rely on heuristics, which can reduce accuracy, particularly in the presence of model misspecification. Moreover, they reconstruct only a single ARG topology and cannot quantify the considerable uncertainty associated with ARG inferences. To address these challenges, we here introduce SINGER, a novel method that accelerates ARG sampling from the posterior distribution by two orders of magnitude, enabling accurate inference and uncertainty quantification for large samples. Through extensive simulations, we demonstrate SINGERs enhanced accuracy and robustness to model misspecification compared to existing methods. We illustrate the utility of SINGER by applying it to African populations within the 1000 Genomes Project, identifying signals of local adaptation and archaic introgression, as well as strong support of trans-species polymorphism and balancing selection in HLA regions.

genetics↗