Po‐Ssu Huang

De novo design of RNA pseudoknots with deep learning

This paper demonstrates that deep learning can accurately design RNA pseudoknots, achieving results comparable to human experts in a competition.

ADAPT-M: A workflow for rapid, quantitative in vitro measurements of enriched protein libraries

ADAPT-M rapidly quantifies protein binder affinities and kinetics using microfluidics, enabling efficient selection of high-affinity variants from yeast-display libraries.

Zero-shot design of a de novo metalloenzyme

This paper introduces dEVA, a framework for de novo metalloenzyme design, achieving high catalytic activity without natural templates.

A generalizable system for antigenic peptide targeting across HLA-I allotypes

The paper develops a system to create binders compatible across HLA allotypes, validated against clinically relevant antigens and demonstrated in CAR T cells.

Leveraging protein representations to explore uncharted fold spaces with generative models

This paper introduces DiffTopo, a diffusion model for generating novel protein folds, and MirrorTopo, a pipeline to expand known fold space by mirroring native protein topologies.

SLAE: Strictly Local All-atom Environment for Protein Representation

SLAE learns all-atom protein representations from local atomic neighborhoods, achieving state-of-the-art performance in downstream tasks through transfer learning.

Isolation of structural constraints by ensemble-conditioned protein sequence design

This paper introduces Caliby, a sequence design method that conditions on structural ensembles to improve designability while reducing native sequence recovery.

Conditional Protein Structure Generation with Protpardelle-1c

Protpardelle-1c is a protein structure generative model that improves motif scaffolding and multi-chain complex generation, achieving higher scores and faster sampling than existing models.

Sidechain conditioning and modeling for full-atom protein sequence design with FAMPNN

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.

Assessing Generative Model Coverage of Protein Structures with SHAPES

This study evaluates generative models of protein structures using SHAPES, revealing undersampling of structural motifs and highlighting the need for better design and prediction methods.

Ig-VAE: Generative Modeling of Immunoglobulin Proteins by Direct 3D Coordinate Generation

The paper introduces Ig-VAE, a variational auto-encoder that generates 3D coordinates of immunoglobulin proteins, demonstrating its use in creating a SARS-CoV2-RBD binder and guiding protein design.

ProGen: Language Modeling for Protein Generation

A 1.2B-parameter autoregressive model trained on 280M tagged sequences completes held-out proteins with near-native Rosetta energies and ranks high-fitness GB1 variants without any GB1 training.

Protein sequence design with a learned potential

This paper designs protein sequences using a deep neural network trained on crystal structures, producing stable designs that match high-resolution structures.

Multi-Scale Structural Analysis of Proteins by Deep Semantic Segmentation

This paper introduces a deep learning method to classify protein structures into architectural classes using semantic segmentation, achieving high accuracy in fold identification and structural quality assessment.