Yun S. Song

CELL-E generates 2D probability density maps of protein localization using text-to-image transformers, avoiding discrete class annotations.

The paper introduces swga2.0, a machine-learning-guided pipeline for designing selective whole genome amplification primers, significantly reducing runtime and improving amplification performance for microbial DNA.

The paper analyzes alternative splicing across diverse mouse cell types using two large-scale single-cell datasets, identifying novel isoforms and splicing regulation patterns.

The paper shows that a single attention layer can predict protein contacts as well as Potts models and pretrained Transformers, suggesting attention models can capture meaningful protein features from unlabeled data.

This paper introduces RNA-Sieve, a likelihood-based method to deconvolve bulk RNA-seq data using single-cell references, providing accurate cell type proportion estimates and confidence intervals.

SCITO-seq enables cost-effective profiling of over 10^5 cells using combinatorial indexing with dsc-seq, demonstrating feasibility in mixed cell lines and human lymphocytes.

The paper uses transporter gene frequencies to predict metabolite characteristics and identifies siderophore-like BGCs in gut microbiomes.

This paper presents a de novo genome assembly of the olive baboon using long-read and Hi-C data, significantly improving contig size and chromosome scaffolding compared to the previous reference genome.

The paper shows Xrn1, a decay factor, modulates RNA polymerase II processivity in yeast genes, affecting transcription initiation, elongation, and termination.

The paper develops a stochastic model of translation dynamics to identify key parameters governing translation efficiency and validates findings with ribosome profiling data from yeast.

Across five protein tasks with evolution-aware splits, self-supervised pretraining helped nearly every model on every task, yet alignment-based features still beat learned embeddings on secondary structure and contact prediction.

The paper analyzes genetic variation in IGHV and TRBV immune receptor gene families across diverse human populations, finding distinct patterns of copy number and mutation.

This paper introduces an exchangeable neural network framework for likelihood-free inference in population genetics, demonstrating superior performance on recombination hotspot testing.

The paper develops ASMC, a fast method to estimate coalescence times from SNP data, detects selection signals and finds enrichment of disease heritability in regions of background selection.

The paper quantifies factors affecting translation elongation rates using ribosome profiling data and shows exit tunnel interactions and nascent polypeptide hydropathy are key determinants.

The paper analyzes the distribution of isolated particles in the TASEP model to estimate mRNA translation rates and identifies potential biases in ribosome profiling data.

The paper introduces MultiCluster, a tensor decomposition method for analyzing multi-tissue gene expression data, showing improved accuracy and speed over existing methods.

This study develops a method to genotype allelic and copy number variation in the immunoglobulin heavy chain locus using whole-genome sequencing data.

This paper develops a Gaussian process method to analyze genomic time series data from experimental evolution, enabling accurate detection and estimation of selection parameters.