Patrick D. Hsu

The paper uses a genomic language model to cluster and annotate noncoding sequences in prokaryotic genomes, identifying structured RNAs and their potential functions.

State, a set-based transformer trained on over 100 million perturbed cells, predicts perturbation responses in new cellular contexts better than mean, linear and deep learning baselines.

A reasoning LLM combining ESM3 embeddings with GO-GPT term predictions produces protein function annotations that experts rated tie-or-better than UniProt in 79% of cases (SFT model).

The paper introduces COMBINE, a platform that screens epigenetic effector domain pairs to identify synergistic and antagonistic interactions, revealing new mechanisms for transcriptional regulation.

ProPer-seq is a cost-effective, barcode-free single-cell perturbation profiling method that enables scalable analysis of synthetic biology perturbations and reveals transcriptional control principles.

This study engineers recombinases to improve DNA insertion efficiency and specificity in human cells, achieving 53% integration and 97% genome-wide specificity.

This paper describes the engineering of IS622 bridge recombinases to enable efficient and specific DNA rearrangements in the human genome, including insertion, inversion, and excision of up to 0.93 megabases.

scBaseCount is a single-cell RNA sequencing database that automates metadata extraction and data processing using AI, offering a large, standardized collection of gene expression data across multiple organisms and tissues.

A 40B-parameter DNA model trained on 9.3 trillion tokens scores coding and noncoding variants zero-shot, exposes interpretable sequence features, and writes chromosome-scale sequences.

This study compares CRISPR-Cas9 and TALEN gene editing in a mouse model of Machado-Joseph disease, showing CRISPR-Cas9 reduces ATXN3 aggregates and delays disease progression.

ESM3, a 98-billion-parameter multimodal protein language model tokenized across sequence, structure, and function, designs a bright green fluorescent protein at 58% sequence identity to known fluorescent proteins.

7B-parameter convolution-attention DNA model with 131 kb context, trained on prokaryotic genomes, performs zero-shot fitness and essentiality prediction and generates Cas loci, transposons, and 650 kb sequences.

This paper introduces RESPLICE, a CRISPR-based trans-splicing system that efficiently edits endogenous human transcripts, achieving up to 90% trans-splicing efficiency in sorted cells.

The paper demonstrates that IS110 insertion sequences use a structured RNA to direct sequence-specific DNA recombination, enabling modular genome editing.

The paper introduces a framework for designing cell-type-specific promoters using model-based optimization, demonstrating improved specificity in blood cancer cell lines.

This paper develops transfer learning strategies to model promoter-driven gene expression in understudied cell types, showing improvements in performance with pretraining and transfer.

The paper uses CRISPR screens to identify mucins as key host factors that restrict SARS-CoV-2 infection in human lung cells.