Dongbo Bu

The paper introduces MultiStateFold, a framework that improves prediction of multiple protein conformations by integrating Parallel Tempering into a protein language model, achieving better performance than existing methods.

ComplexDesign is a method for designing multichain protein complexes by optimizing sequences to fold proteins and form inter-chain interactions, achieving high success rates in designing dimers, trimers, and tetramers.

This paper introduces OCDesign, a curriculum-guided framework for multi-property protein design, showing that staged objective introduction improves success rates compared to one-shot optimization.

This paper introduces a modular DNA storage system using dynamic DNA bytes that enables reconfigurable, rewritable data storage with CRUD operations and error correction.

This paper introduces ProTCR, a reinforcement learning method for predicting TCR-pMHC binding, achieving higher accuracy than existing methods on benchmark datasets.

ProAffinity++ predicts protein-protein binding affinities using sequence and structure representations, outperforming existing methods in accuracy and handling complex cases like antigen-antibody interactions.

This paper introduces NeuralNJ, a deep learning method for phylogenetic inference that improves accuracy and efficiency by using an end-to-end framework and reinforcement learning.

This paper introduces a cost-effective DNA storage system using pre-fabricated DNA movable types that can be reused, reducing storage costs to $122/MB.

The paper introduces ImmuneFold, a transfer learning method for immune protein structure prediction, showing improved accuracy over existing methods for TCR-epitope binding.

RNA-km, a language model, predicts RNA secondary structures including pseudoknots without labeled data, showing strong generalization on benchmark datasets.

CarbonDesign improves protein sequence design by integrating AlphaFold concepts, achieving high accuracy on benchmark datasets and showing robustness in de novo design and variant effect prediction.

ProDESIGN-LE designs protein sequences efficiently by learning residue local environments, achieving high accuracy in structure prediction and experimental validation.

KnotFold predicts RNA secondary structures including pseudoknots with higher accuracy than existing methods by combining learned potentials with a minimum-cost flow algorithm.

This study introduces DeepNovo Peptidome and DeepSelf for neoantigen discovery and validation using immunopeptidomics and deep learning, demonstrating their application in cervical cancer tissues.

ProALIGN improves protein structure prediction by learning optimal sequence-template alignments using deep learning, outperforming existing methods on multiple datasets.

CopulaNet directly learns residue co-evolution from multiple sequence alignments to improve protein structure prediction accuracy and efficiency.

The paper uses bi-clustering to identify gene clusters with consistent expression-protein abundance correlations and builds a prediction model to forecast protein abundance from mRNA data.