Yun S. Song

Replaying germinal center evolution on a quantified affinity landscape

Across 119 monoclonal germinal centers, B cell affinity rises reliably despite divergent phylogenies. Selection is noisy but persistent; mutability limits exploration. A DMS-based landscape shows saturating fitness and survivorship biases.

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

NovaClone improves clonal and subclonal genotyping in single-cell lineage tracing by addressing technical artifacts and providing quality control metrics.

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

The paper uses simulation-based deep learning to infer the affinity-fitness response function in germinal centers from experimental data.

Separating selection from mutation in antibody language models

This paper introduces a model that separates mutation and selection processes in antibody language models to improve prediction of mutation effects on antibody function.

The Impact of Stability Considerations on Genetic Fine-Mapping

This paper introduces a stability-guided fine-mapping approach that complements residualization, showing it enhances functional variant identification in eQTL analysis.

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

PEINT simulates realistic protein evolution by modeling sequence changes over time, generating functional proteins that retain enzymatic activity.

Predicting the effect of CRISPR-Cas9-based epigenome editing

The paper trains machine learning models to predict gene expression from histone PTMs using ENCODE data and validates predictions with CRISPR-Cas9-based epigenome editing experiments.

Architecting cis-regulation to quantitatively tune gene expression in cereals

The paper develops a high-throughput MPRA to map cis-regulatory elements in cereal promoters, identifying mutations that significantly increase gene expression.

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

This study compares nucleotide context models and protein language models for predicting antibody affinity maturation, finding nucleotide-based models perform better.

A General Framework for Branch Length Estimation in Ancestral Recombination Graphs

This paper introduces POLEGON, a method for estimating branch lengths in Ancestral Recombination Graphs using uninformative priors, showing improved coalescence time estimates and population size inferences.

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

This paper introduces GPN-Star, a phylogeny-informed genomic language model that improves variant effect prediction across species and evolutionary timescales.

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

This paper introduces QRscore, a non-parametric test for detecting distributional shifts in gene expression, showing it outperforms existing methods in detecting variance and mean shifts.

ConvexML: Fast and accurate branch length estimation under irreversible mutation models, illustrated through applications to CRISPR/Cas9-based lineage tracing

This paper introduces ConvexML, a fast and accurate method for estimating branch lengths in phylogenetic trees under irreversible mutation models, with applications to CRISPR/Cas9 lineage tracing data.

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

This work benchmarks DNA sequence models for predicting causal regulatory variants in human genetics using a curated dataset of 113 Mendelian and 83 complex traits.

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

CPI-Pred predicts compound-protein interaction parameters using deep learning, outperforming existing models on unseen compounds and enzymes.

GPN-MSA: an alignment-based DNA language model for genome-wide variant effect prediction

GPN-MSA is a DNA language model that predicts variant effects across the human genome using whole-genome alignments, achieving strong performance on deleteriousness prediction for coding and non-coding variants.

A previously reported bottleneck in human ancestry 900 kya is likely a statistical artifact

The paper challenges a previously reported 900 kya bottleneck in African ancestry, arguing it is a statistical artifact rather than a true demographic event.

Learning antibody sequence constraints from allelic inclusion

The paper uses allelic inclusion data to train models that identify antibody sequence constraints, finding stronger correlations with antibody properties than existing methods.

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

This paper develops a Neighbor-Joining algorithm with theoretical guarantees for reconstructing phylogenies from CRISPR-Cas9 lineage tracing data, showing improved performance on simulated and real data.

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

This study uses deep scanning mutagenesis to analyze POT1 mutations in human stem cells, revealing that some cancer-associated variants drive telomere elongation and oncogenesis independently of telomere replication.

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

SINGER improves ARG reconstruction accuracy and uncertainty quantification for large genomic samples by enabling efficient posterior sampling.

Highly parameterized polygenic scores tend to overfit to population stratification via random effects

The paper shows that highly parameterized polygenic scores overfit to population stratification by including random variants, reducing their generalizability.

Tracing cancer evolution and heterogeneity using Hi-C

The paper develops HiDENSEC to analyze Hi-C data in heterogeneous cancer samples, identifying chromosomal rearrangements and their impact on melanoma progression.

DNA language models are powerful predictors of genome-wide variant effects

This paper introduces GPN, a DNA language model that predicts genome-wide variant effects and outperforms conservation-based predictors in Arabidopsis thaliana.

Cross-protein transfer learning substantially improves disease variant prediction

This paper introduces a cross-protein transfer learning framework that improves disease variant prediction using deep mutational scanning data from few proteins, achieving better specificity than existing models.

Reference-free multiplexed single-cell sequencing identifies genetic modifiers of the human immune response

The paper introduces clue, a reference-free framework for multiplexed single-cell sequencing to identify genetic modifiers of immune responses using SNP-based clustering and eQTL mapping.

The ENCODE Imputation Challenge: A critical assessment of methods for cross-cell type imputation of epigenomic profiles

The paper evaluates 23 imputation models for cross-cell type epigenomic data, finding quality assessment challenging due to distributional shifts, data availability, and redundant metrics.

CELL-E 2: Translating Proteins to Pictures and Back with a Bidirectional Text-to-Image Transformer

CELL-E 2 translates protein sequences to images of subcellular localization and vice versa, enabling de novo protein design and generating novel nuclear localization signals.

Tuned Fitness Landscapes for Benchmarking Model-Guided Protein Design

This paper introduces SLIP, a framework for creating synthetic protein fitness landscapes to benchmark ML-guided protein design pipelines.

The Impact of Protein Dynamics on Residue-Residue Coevolution and Contact Prediction

The paper shows that protein dynamics impose evolutionary constraints and that unsupervised language models can predict these constraints from sequence data.