Thu NguyenPh.D.

About

Thu Nguyen, Ph.D.

I’m a computational scientist trained in computer science and shaped by structural biology. I enjoy problems where a model must do more than perform well on a benchmark: it must respect scientific constraints, scale to real research workflows, and produce results that domain experts can interrogate.

My path has taken me from machine learning and scientific data analysis in Vietnam to crystallography at Brookhaven National Laboratory, doctoral research in protein structure prediction at Stony Brook University, and computational protein design in industry. Across those settings, I have built models, reproduced complex research systems, extended neural architectures, and made training workflows usable across GPU and cloud environments.

I’m especially interested in the next generation of AI methods for biological design: models that combine structure, sequence, evolutionary context, physical constraints, and experimentally meaningful objectives.

Abstract protein-inspired composition.

Capabilities

What I bring to a research team

Scientific ML

TransformersGNNsCNNsLSTMs / RNNsDiffusionFlow matchingGeometric deep learning

Protein AI

AlphaFoldESMRFdiffusion / RFD3ProteinMPNNLigandMPNNFAMPNNSaProtADFLIPBioEmuPoET

Engineering

PythonPyTorchTensorFlowGCPCUDA / ROCm workflowsRJavaC/C++

Research domains

Computational biologyProtein designDrug discoveryStructural biologyCrystallography

Education

Education

Stony Brook University
Ph.D., Computer Science
Stony Brook, NY · December 2023
Vietnam National University
B.S., Computer & Information Science, High Distinction
Hanoi, Vietnam · June 2018

Career

Full career history

  1. Dec 2023 – Present

    Insmed Inc.

    Scientist

    Computational protein design and deimmunization: sequence optimization, inverse folding, transformer-based fitness prediction, pH-sensitive binder design, and cloud-based scientific computing.

    • Implemented MCMC with NetMHC score and IF1/ProteinMPNN score to design and deimmunize protein sequences. Achieved more than 70% NetMHC score reduction and a 10% TM/activity increment on monomer Griffithsin.
    • Implemented property guidance to ADFLIP to design deimmunized proteins, achieving a 40% NetMHC score reduction on SaCas9.
    • Designing and training a transformer-based model that integrates templates and MSAs to design protein sequences with fitness-score prediction. The model achieved sequence recovery up to more than 60% on validation data, versus roughly 45-50% for IF-1 and ProteinMPNN; its fitness-score correlation with Ides activity reached up to 0.7, compared with 0.5 for ESM-IF1.
    • Designing pH-sensitive binders (nanobodies and ligands) using LigandMPNN and RFD3 for Pdgfr-b.
    • Set up folding models on GCP for internal users; pushed training workflows to GCP Batch and Vertex AI.

  2. Aug 2020 – Dec 2023

    Stony Brook University, Applied Bio Computation Lab

    Research Assistant

    Doctoral research in geometric deep learning and structure prediction, including AlphaFold reproduction and SE(3)-Transformer extensions.

    • Implemented Generalized Born (GB) scoring for FTMap, improving docking accuracy by 10% based on RMSD metrics.
    • Improved SE(3)-Transformer by adding extra self-attention kernels; the new model outperformed the original model's precision by up to 30% on an n-body dataset.
    • Reproduced monomer and multimer models from AlphaFold for training in PyTorch, compatible with ROCm and CUDA and trainable across thousands of nodes and GPUs.
    • Fine-tuned an AlphaFold model by adding network layers for better prediction of CDR loops, antibody-antigen interactions, and new residue positions.
    • Implemented a deep network based on AlphaFold to design a peptide sequence whose structure binds to a desired location.

  3. Jun 2019 – Jun 2021

    Brookhaven National Laboratory

    Research Assistant

    Crystallography data processing: diffraction-image classification, lossy compression, and structure-factor analysis.

    • Implemented lossy compression for X-ray diffraction images to reduce memory storage.
    • Developed a convolutional neural network (CNN) in PyTorch to detect and classify indexable diffraction images.
    • Applied structure-factor clustering to detect movement behavior in binding hotspots, revealing small structural differences in chymotrypsinogen.

  4. Sep 2018 – Dec 2018

    Stony Brook University, Computational Bio Lab

    Research Assistant

    Sequence alignment performance and parallelization.

    • Enhanced sequence alignment performance using matching constraints and cached alignment results in C++.
    • Implemented parallel sequence alignment using OpenMP for improved efficiency.

  5. Feb 2016 – Aug 2018

    Vietnam National University, University of Science

    Research Assistant

    Early research in applied machine learning, NLP, and statistical forecasting.

    • Built a machine learning model to classify sentiment polarity on restaurant reviews in Python.
    • Developed the first Vietnamese semantic network for medical data by cleaning data, extracting collocations, constructing lexicons, and clustering documents using Java and Python.
    • Analyzed rainfall data from 12 rain gauge stations (2008-2016) to forecast landslides in Central Vietnam using ARIMA and Bayesian models in R and Python.
Get in touch