KennethZheng

Berkeley microbial biology student exploring machine learning, biotech, and design.

Explore my work
A three-dimensional knotted surface composed of tiny A, C, G, and T letters

I study microbes and build computational tools for biology, from antimicrobial resistance models to interactive learning.

Selected work

Visual experiments

Original interface studies with synthetic data and mathematical artwork.

Synthetic cluster explorerGenerative flow field

Microbial biology at Berkeley.

I study Microbial Biology at UC Berkeley, with a minor in Data Science and expected graduation in May 2027. Lab experience informs how I approach biological data, model evaluation, and the interfaces people use to understand them.

  • LanguagesEnglish, Mandarin, Cantonese
  • Off screenWushu & violin

Experience

Clinical Laboratory Scientist Intern

Guangdong Kanghua Hospital

PCR/qPCR, antimicrobial susceptibility assays, and quality-controlled clinical laboratory workflows.

June to July 2026

Youth Program Instructor

UC Berkeley Recreation & Wellbeing

Designed educational and recreational programs for children and adolescents.

May to August 2025

Screening Volunteer

Kidney Disease Screening & Awareness Program

Supported community health screenings and Mandarin/Cantonese interpretation.

2024 to 2026

Biology, connected.

BASE brings biological entities, relationships, and sources into one explorable archive.

Founder & developer / Biological education / In beta

BASE’s live human metabolism page with pathway navigation and learning cards

Follow the relationships.

Biological Archive for Science & Education puts structured records and their connections before isolated articles. Metabolism is one entry point: a molecule leads to a reaction, an enzyme, and the pathway around it.

Keep the evidence close.

Curated explanations sit alongside attributed sources and provenance. The design question is how to keep scientific depth accessible without making the interface feel like a database.

Visit BASE

AMR Phenotype Predictor

A research prototype for classifying carbapenem resistance in Acinetobacter baumannii.

Independent project / BV-BRC / Python & XGBoost

Resistance-gene features feed into model comparison and resistance classification

From genomic features to prediction.

Compared logistic regression, random forest, and XGBoost using resistance-gene presence/absence features from 3,999 BV-BRC genomes.

Evaluation

The project README reports XGBoost ROC-AUC of 0.992 under stratified five-fold cross-validation. This result has not been established on an independent test set or in clinical use.

View the project

Single-Cell PBMC Walkthrough

Exploring immune-cell identities in the public PBMC3k dataset.

Independent project / Single-cell analysis / Scanpy

Scanpy analysis followed by marker-based annotation of immune-cell types

From expression to cell types.

A Scanpy workflow for quality control, normalization, neighbourhood graphs, UMAP, Leiden clustering, and annotation with canonical marker genes.

Grounding interpretation in markers.

The walkthrough examines monocytes, B cells, T cells, and NK cells. Its focus is a transparent analysis workflow for a public reference dataset.

View the walkthrough

RNA-seq under antibiotic exposure

Reanalysis of E. coli gene expression in a public nine-sample dataset.

Public-data reanalysis / PRJNA525186 / PyDESeq2

FastQC and MultiQC, Salmon quantification, then PyDESeq2 differential expression analysis

Reads to differential expression.

The dataset contains three conditions with three replicates each. The pipeline uses FastQC/MultiQC for quality control, Salmon for quantification, and PyDESeq2 for differential expression.

Reported findings

The project README reports 329 and 300 significant genes across two antibiotic-exposure contrasts. The repository documents the analysis approach; its end-to-end reproduction instructions remain incomplete.

View the analysis

Biological signals

An interface experiment for exploring synthetic data clusters.

Interactive study / Synthetic data / No trained model

Separation: 65%. Synthetic data.

From data to something you can question.

Move the slider to change how clearly the groups separate. This is a visual study of an exploration interface, not an analysis of biological samples or a validated model.

Living patterns

An exploration of organic form through simple mathematical rules.

Generative study / Parametric flow field

Complexity: 42. Generative artwork.

Rules become form.

A field of curves grows from a shared function. Small changes produce different folds, densities, and rhythms. It is a study in emergence and visual systems, not a simulation of a specific organism.