Kyunghyun Cho

GLM-Prior is a genomic language model that generates sequence-derived priors for gene regulatory network inference, showing strong performance in mammals but limited success in yeast.

DyAb predicts antibody properties and designs new sequences with high binding affinity using a low-data approach.

ConformAb designs antibody sequences that maintain CDR canonical conformations, improving binding affinity without target-specific data.

Property Enhancer optimizes antibody properties using a data-efficient framework, achieving significant affinity gains in both in silico and in vitro tests.

The paper compresses protein sequence and structure representations, showing that compressed embeddings retain structural accuracy and enable downstream applications.

The paper introduces a segmentation-free vision transformer, CANVAS, to discover and characterize tumor microenvironment signatures from high-dimensional multiplexed images, identifying a monocytic signature linked to poor prognosis.

This paper introduces a machine learning framework for therapeutic antibody design that integrates computational methods with experimental validation, resulting in significantly improved binding affinity for multiple targets.

PLAID generates controllable all-atom protein structures from sequences using a diffusion model trained on sequence data, enabling functional control through Gene Ontology and organism keywords.

This paper trains a protein language model in 24 GPU hours, achieving performance comparable to a much larger model trained for 15,000x more GPU hours.

The paper introduces a framework using protein language models to learn sequence-structure-function representations and develops a retrieval system for large-scale protein annotation and function prediction.

This paper introduces a neural machine translation model to generate natural language descriptions of protein functions, enabling the discovery of novel functions beyond existing label spaces.

EquiFold predicts all-atom protein structures using an SE(3)-equivariant coarse-grained geometric representation without sequence alignments or language model embeddings, matching AlphaFold accuracy orders of magnitude faster in preliminary tests.

This paper introduces PMF-GRN, a variational inference method for single-cell GRN inference, showing improved accuracy and uncertainty estimates compared to existing methods.

NetTIME improves TF binding site prediction using multitask learning and base-pair resolution, outperforming existing methods in supervised and transfer learning settings.

TM-Vec predicts structural similarity from sequence pairs to detect remote homologs, outperforming sequence alignment methods and matching structure-based alignment performance.

This paper introduces a deep learning method for protein design that uses a sequence denoising autoencoder and function predictor to guide sampling towards functional sequences.

This paper introduces a Graph Convolutional Network for predicting protein function and identifying important residues, showing improved accuracy over sequence-based methods.

NetQuilt improves protein function prediction across species by integrating sequence and network data from multiple organisms.

This paper introduces DEWAKSS, a method for denoising single-cell genomics data by optimally tuning kNN- and diffusion-based approaches to preserve biological variance and improve clustering.