Brian Hie

This paper demonstrates that a genomic language model can generate functional regulatory DNA sequences using few examples, showing improved performance in cell-type-specific gene expression and neuronal labeling.

This paper demonstrates that in-context learning in the genome language model Evo 2 enables efficient biological sequence design without additional training.

This study uses machine learning to guide directed evolution of plant Rubisco, identifying variants with improved catalytic efficiency and carboxylation rates.

This paper designs a new proximity labeling enzyme, FlexID, using computational methods to improve spatial precision and versatility for mapping protein interactions and proteomes.

Minerva extracts RNA base-pairing, protein-contact and repeat maps from a genome language model without alignments, flagging unannotated structured loci and a phage ncRNA array reverse-transcribed into short cDNAs.

This paper introduces antiGen, a machine learning model that forecasts viral mutations from phylogenetic trees, achieving state-of-the-art results for SARS-CoV-2 and other viruses.

This paper aligns a protein generative model with experimental fitness using Direct Preference Optimization, improving stability prediction and enabling new applications.

This paper introduces Proto, a high-level programming language for generative biology, and demonstrates its use in designing and validating biological systems.

This paper introduces a genomic language model, Evo, capable of generating functional de novo genes through semantic mining, validated by experimental testing of anti-CRISPR proteins and toxin-antitoxin systems.

Germinal is a generative pipeline that designs functional antibodies with nanomolar binding affinities using minimal experimental testing.

Fine-tuned Evo 1 and Evo 2 genome language models designed complete ΦX174-like phage genomes; 16 of 285 assembled designs were viable, and some outperformed ΦX174 in competition.

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 paper introduces FAMPNN, a method for protein sequence design that explicitly models both amino acid identity and sidechain conformation, achieving state-of-the-art results in sidechain packing and practical design applications.

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.

The paper uses protein language models to predict allosteric residues from single sequences without structural data, outperforming existing methods on benchmark sets.

This paper uses a structure-informed language model for unsupervised antibody evolution, achieving significant improvements in neutralization and affinity against viral variants.

Consensus likelihoods from general, unsupervised protein language models evolve seven human antibodies using only a wildtype sequence and <=20 tested variants, improving binding affinity up to 160-fold.

Protein language models scaled to 15B parameters predict atomic-resolution structures from sequence up to 60x faster than the state of the art, producing over 617 million metagenomic structures.

A modular high-level programming language paired with an energy-based protein language model generates de novo all-atom protein sequences and structures satisfying user-specified geometric, secondary structure, symmetry, and multimerization constraints.

Adding 12 million AlphaFold2-predicted structures to training lifts native sequence recovery on held-out backbones to 51.6%, versus 42.2% for the best experimental-only model; likelihoods score mutation effects zero-shot.

This paper introduces evolutionary velocity, a method using protein language models to predict evolutionary order and direction, showing its effectiveness across different timescales and providing new insights into viral-host immune escape and horizontal gene transfer.

The paper organizes a DREAM challenge to benchmark predictive models of single-cell signaling using mass cytometry data from breast cancer cell lines.

The paper introduces Schema, a metric learning algorithm that synthesizes heterogeneous single-cell modalities to infer cell types, analyze gene expression, and estimate evolutionary pressure.

The paper uses neural language models to predict viral escape mutations by modeling semantic change and grammaticality in viral sequences.

The paper demonstrates how uncertainty quantification in machine learning improves biological discovery by enabling robust predictions and iterative experimental design across diverse domains.

The paper introduces Trajectorama, an algorithm that uses gene coexpression to integrate cellular trajectories across studies, revealing developmental and disease-related processes.

The paper introduces a geometric sketching method to compactly summarize scRNA-seq data, revealing rare cell types and accelerating analysis tasks.

Scanorama matches mutual nearest neighbors across all scRNA-seq data set pairs, rapidly merging 105,476 cells from 26 experiments while keeping data set-specific cell types separate.

This study fine-maps cis-regulatory variants across diverse human populations by analyzing chromatin accessibility QTLs and their effects on gene expression and long-range interactions.