Banghua's personal website

Banghua Zhu at



Research Project


My research interest includes computational biology, machine learning and information theory.

Hybrid Assembly of Long Tandem Repeats on Mucin 2 Gene

Stanford University, CA, USA 2017.7 - 2017.9

Electrical Engineering Department

Research Assistant, Advisor: Prof. David Tse & Prof. Andrew Fire

· Cleaned and processed data for Pacbio reads, Illumina reads and Nanopore reads.

· Designed assembly pipeline to deal with tandem repeats.

· Successfully assembled reliable Mucin 2 gene, which is related to colon cancer.

· Developing tools for assembling tandem repeats.

· See project poster here.

Haplotype Assembly via Community Detection

Stanford University, CA, USA 2017.7 - 2017.9

Electrical Engineering Department

Research Assistant, Advisor: Prof. David Tse

· Applied spectral stitching algorithm to haplotype assembly problem. (Implemented the algorithm in Paper (ICML), which is a previous work in Tse’s lab.)

· See Project website for more information.

Haplotype Assembly via Tensor Factorization

The University of Texas at Austin, Austin, TX, USA 2016.8 - 2016.12 Electrical and Computer Engineering Department

Research Assistant, Advisor: Prof. Haris Vikalo

· Applied tensor factorization framework to haplotype assembly problem.

· Helped establish theoretic bounds in error under certain coverage condition.

· Implemented algorithms for alternative gradient descent in tensor factorization in both matlab and python.

· Experimented on 1000 genome data, fosmid data and simulation data.

· Paper accepted by RECOMB-seq: A Tensor Factorization Framework for Haplotype Assembly of Diploids and Polyploids

· Paper to be published by BMC Genomics: Sparse Tensor Decomposition for Haplotype Assembly

Image denoising by low dimensional manifold model

Tsinghua University, Beijing, China 2017.1 - present

Yau Mathematical Sciences Center

Research Assistant, Advisor: Prof. Zuoqiang Shi

· Used low dimensional manifold model to denoise images.

· Demonstrated the superiority of low dimensional manifold model compared to low rank model.

· Applied Weighted Nonlocal Laplacian Method (WNLL) and Point Integral Method (PIM) to solve a Laplace-Beltrami equation for low dimensional manifold model.

· Paper in edit: Low dimensional Manifold Model for Impulsive Noise Removal

Decision Tree and Information Theory

Qingfan MOOP / Stanford University, Stanford, CA, USA 2016.11 - present Department of Electrical Engineering

Intern, Advisor: Prof. Tsachy Weissman, Jiantao Jiao

· Applied JVHW estimator to decision tree and image registration problems.

· Compared 12 entropy estimators to demonstrate the superiority of JVHW estimator.

· Provided theoretical guarantee and empirical evidence to show the increase of performance with the application of JVHW estimator.

· Paper to be submitted to ICML: Boosting Decision Tree Learning by Optimal Split Criterion Estimation. Manuscript

Promoter-Enhancer Interaction Prediction

Tsinghua University, Beijing, China 2016.8 - present

Institute for Interdisciplinary Information Sciences

Research Assistant, Advisor: Prof. Jianyang Zeng

· Designed a deep neural network framework for promoter-enhancer interaction prediction.

· Experimented on Capture Hi-C data and PHiC data.

· Analyzed derived patterns from learning.

Spatial Modulation and Sparse Bayesian Learning

Tsinghua University, Beijing, China 2015.9 - 2016.8

Wireless Communication Group, Department of Electronic Engineering

Research Assistant, Advisor: Prof. Jintao Wang

· Applied Sparse Bayesian Learning method for Spatial Modulation.

· Improved Sparse Bayesian Learning to utilize the integer property in symbol detection.

· Paper Accepted by IEEE Broadcasting Symposium: Fast Sparse Bayesian Learning based Symbol Detection for Massive Spatial Modulation. Oral presentation on IEEE Broadcasting Symposium 2016.