Undergraduate in Data Science, UC Berkeley

Qiankang (Kant) Wang

My research focuses on deep learning for DNA and protein sequences, and on which parts of known biology a model should be given rather than left to learn.

Scroll

About

Deep learning models can now learn directly from DNA and protein sequences, without labels, and generate new ones. I want to work out how architectures, training objectives, and what we already know about biology can help self-supervised and generative models learn what we don't.

  • Machine learning for DNA and protein sequences
  • Self-supervised and generative models
  • Inductive biases from known biology

At Stanford, I compared two deep learning models of gene regulation, AlphaGenome and ChromBPNet, on equal footing.

Before that, I worked with Prof. Ray Luo at UC Irvine, starting while I was still at Diablo Valley College, a community college. I wrote iterative solvers for AmberTorchPB, a tensor-based framework for molecular electrostatics, and that work is now published in the Journal of Chemical Theory and Computation.

Research

  1. Jul 2024–Feb 2026

    Scientific computing · GPU tensor computation

    Iterative Poisson–Boltzmann Solvers for AmberTorchPB

    University of California, Irvine

    Implemented conjugate gradient (CG) and biconjugate gradient (BiCG) solvers in AmberTorchPB, a LibTorch framework for Poisson–Boltzmann electrostatics on CPUs and GPUs, and built a Slurm pipeline that ran more than a million Poisson–Boltzmann surface area (PBSA) calculations to benchmark accuracy, runtime, and memory across precisions.

    Details
    • Implemented conjugate gradient (CG) and biconjugate gradient solvers in AmberTorchPB, a LibTorch framework for Poisson–Boltzmann reaction-field energies on CPU and GPU.
    • Built a Slurm pipeline that ran over one million Poisson–Boltzmann surface area (PBSA) energy calculations to benchmark accuracy, runtime, and memory across precisions; under identical settings, the CG solver matched AMBER PBSA energies (R² = 1.00) while running over 2× faster on CPU and nearly 2× faster on GPU.
  2. May 2026–Oct 2026

    Deep learning · Sequence-to-function models

    Sequence-to-Function Models: AlphaGenome and ChromBPNet

    Stanford University

    For a fair comparison on identical held-out K562 regions, retrained ChromBPNet on AlphaGenome’s data split and aligned input windows and evaluation loci. Evaluated layer embeddings from both models with linear probes and UMAP under a shared, overlap-aware cross-validation protocol, and applied the same attribution and motif analysis to both.

    Codegithub.com/wangkant/motifline (opens in new tab)

    Details
    • Retrained ChromBPNet on AlphaGenome’s data split, aligning input windows and evaluation loci to enable a fair comparison on identical held-out K562 regions.
    • Probed layer embeddings from AlphaGenome and ChromBPNet with linear probes and UMAP under a shared, overlap-aware cross-validation protocol, testing which results hold across embedding readouts.
    • Built an open-source pipeline that applies identical attribution (DeepLIFT/DeepSHAP), motif discovery (TF-MoDISco), and hit calling (FiNeMo) to AlphaGenome and ChromBPNet.

Publications

  1. AmberTorchPB: A Unified Framework for Poisson–Boltzmann-Based Reaction Field Energy Calculation via Tensor Computation (opens in new tab)

    Wu, Y.; Wang, Q.; Jiang, R.; Luo, R. Journal of Chemical Theory and Computation 2026, 22(7), 3554–3570. Second author

    A LibTorch framework for Poisson–Boltzmann reaction field energies on CPUs and GPUs. Under identical settings, its CG solver matches AMBER PBSA energies (R² = 1.00) while running over 2× faster on CPU and nearly 2× faster on GPU.

    My contributionImplemented the CG and BiCG solvers and benchmarked accuracy, runtime, and memory use across numerical precisions.

    Paper (opens in new tab)
    BibTeX
    @article{wu2026ambertorchpb,
      title   = {{AmberTorchPB}: A Unified Framework for {Poisson--Boltzmann}-Based Reaction Field Energy Calculation via Tensor Computation},
      author  = {Wu, Y. and Wang, Q. and Jiang, R. and Luo, R.},
      journal = {Journal of Chemical Theory and Computation},
      year    = {2026},
      volume  = {22},
      number  = {7},
      pages   = {3554--3570},
      doi     = {10.1021/acs.jctc.6c00085}
    }

Projects

Education

  1. Aug 2025–May 2027

    University of California, Berkeley

    B.A. in Data Science

    GPA 3.82 / 4.00

    Coursework Introduction to Machine Learning (COMPSCI 189) · Natural Language Processing (EECS 183) · Concepts of Probability (STAT 134) · Introduction to Computational Molecular and Cell Biology (BIOENG C131)

    Honors

    • Data Science Honors Program, University of California, BerkeleyAug 2026–Present
    • Dean’s List, UC Berkeley College of Computing, Data Science, and SocietyDec 2025
  2. Jan 2023–May 2025

    Diablo Valley College

    Data Science

    GPA 4.00 / 4.00

    Coursework Linear Algebra (MATH 194) · Differential Equations (MATH 294) · Elementary Statistics with Probability (MATH 142) · Introduction to Data Science (COMSC 156) · Program Design and Data Structures (COMSC 210)

News

  1. Joined the Data Science Honors Program at UC Berkeley.

  2. Started research at Stanford comparing AlphaGenome and ChromBPNet.

  3. AmberTorchPB paper published in the Journal of Chemical Theory and Computation (opens in new tab).

  4. Named to the Dean’s List at UC Berkeley.

  5. Joined Ray Luo’s lab at UC Irvine to work on Poisson–Boltzmann solvers.