Ask better questions. Then teach the answers.
7 papers across genomics, the gut microbiome and biochemistry, from the wet lab to interpretable machine learning, and a decade of teaching people to think with data.
An Interpretable DNA Methylation Framework for Robust Glioma Classification and Survival Modeling
97.4%accuracy on a held-out test set (113/116)
93–100%accuracy across external 450k & EPIC cohorts
404K → 70probes to an interpretable CpG signature
An open-source, leakage-controlled DNA methylation pipeline that classifies diffuse glioma subtypes and explores survival risk on interpretable gene-region features.
Publications
The gapMethylation-based diagnostic tools for glioma often act as black boxes, and many prognostic models come from single cohorts and are prone to overfitting, which limits trust and generalizability.
My contributionFirst author. Built the leakage-controlled pipeline on Illumina 450k and EPIC data: a multi-class XGBoost classifier reduced from 403,954 probes to 70 CpGs, and penalized Cox models (pooled cancer-type stratified Elastic-Net, plus per-histology Elastic-Net and CoxBoost) on gene-region features under nested cross-validation.
Outcome97.4% accuracy on a held-out test set never used for tuning, and 93–100% across external 450k and EPIC cohorts. Survival modelling is exploratory: within-histology concordance of 0.59–0.72, with candidate prognostic loci including HMGA2, RHOBTB3, HSPG2 and TRMT61B that need external clinical validation. SSRN preprint, 17 pages; under peer review.
The gapStandard tumor classification relies on histology, but ~10–20% of tumors are misclassified — sending patients to the wrong treatment regimen with major survival consequences.
My contributionBuilt the multimodal ML pipeline integrating bulk RNA-seq gene expression with DNA methylation features. Engineered the joint feature space, trained ensemble classifiers, and developed the misclassification-flagging logic that surfaces tumors where the molecular signal disagrees with histological label.
OutcomePipeline flags histologically misclassified tumors that would otherwise be treated incorrectly. Published in Electronics (2025) as evidence that integrating multi-omics layers can correct the diagnostic gap that single-modality classifiers miss.
The gapIndian immigrants to Western countries develop higher rates of inflammatory bowel disease and metabolic disorders within a generation. The gut microbiome is suspected as a mediator, but the exact compositional shifts and their dietary drivers were unmapped.
My contributionCo-led the gut microbiome analysis and visualization pipeline in Python and R. Designed the comparative analysis between Indian residents, recent immigrants, and second-generation Indo-Canadians. Built the visualization layer that mapped specific dietary acculturation patterns to compositional shifts in gut microbiota.
OutcomePublished in npj Biofilms and Microbiomes (2025). Demonstrated that westernization of the gut microbiome tracks dietary acculturation rather than time-since-immigration alone — implicating modifiable diet patterns as the primary intervention target.
The gapMigration-driven changes in the gut microbiome were known qualitatively but not statistically validated across pre- and post-migration populations with matched comparisons.
My contributionDesigned and executed the statistical comparison between Indian-resident and Indo-Canadian gut microbial communities. Selected appropriate non-parametric tests for compositional data, controlled for confounders (age, diet diversity), and quantified effect sizes.
OutcomeConference paper in the Journal of the Canadian Association of Gastroenterology (2024). Set the statistical foundation for the subsequent npj Biofilms paper that extended this analysis to dietary acculturation drivers.
The gapDiet shapes the gut microbiome, but how dietary lipid composition (saturated vs. unsaturated) affects gut mucosal proteins — the host's biochemical interface with microbes — was poorly characterized.
My contributionFirst-author study. Designed and ran the bioinformatics workflow integrating gut mucosal proteomics with 16S microbiome sequencing across mice on different lipid-saturation diets. Built the multi-omics analysis pipeline from raw reads through differential abundance to functional annotation.
OutcomePublished in Nutrients (2019). Demonstrated that the host's gut mucosal proteome and the resident bacteriome are shaped by dietary lipid saturation index — implicating fatty-acid quality (not just quantity) in gut-immune signaling.
The gapGeographic location strongly correlates with gut microbiome composition, but the role of local environmental drivers (vs. genetic, dietary) remained unclear in the literature.
My contributionContributed the environmental-driver statistical analysis in R. Tested associations between local environmental variables and gut microbial diversity across geographically distinct populations.
OutcomePublished in Frontiers in Microbiology (2017). Provided statistical evidence that environment is a significant independent driver of gut microbiome variation, beyond diet and host genetics.
The gapStandard nutritional guidance counts calories, but enteric infection severity may depend on the type of dietary lipid — a question that requires controlled animal experiments and proteomic readout.
My contributionRan the gut mucosal proteomic experiments and contributed to downstream data analysis. Performed wet-lab proteomic workflow (sample prep, MS) and analyzed differentially expressed proteins across diet groups in infected mice.
OutcomePublished in The Journal of Infectious Diseases (2016). Showed that lipid type — not total caloric intake — alters infection outcomes in a mouse model of enteric infection, with implications for dietary recommendations during gastrointestinal illness.
Peer review
On the other side of the review.
I peer review manuscripts for scientific journals. Review credits are verified on ORCID.
Dynamic Association between Gut Microbiota Succession and Host Metabolism: Differential Effects of Antibiotic Intervention Duration on Glucose and Lipid Metabolism in High-Fat Diet-Induced Obese Mice
How the duration of antibiotic exposure shapes gut microbiota recovery, short-chain fatty acids and metabolic health in obese mice.
Review credit on ORCID ↗
Teaching
The best way to learn it is to teach it.
From graduate data-science courses to one-on-one mentoring and 60-second lessons, I teach the way I wish I’d been taught: real data, clear reasoning, no rote syntax.
37MDS students
Sessional Lecturer
UBC Okanagan · Master of Data Science
Taught Programming in Python and R to 37 MDS students, built around reproducibility and clean code rather than rote syntax.
30+researchers trained
Research Data Consultant
UBC Okanagan
Designed and ran Python and R workshops for researchers across departments, plus one-on-one help on real datasets.
11courses supported
Teaching Assistant
UBC Okanagan · Biology, Chemistry, Physics, Biochemistry
Led labs, tutorials and exam reviews across the first- and second-year STEM core, then graduate-level biology courses.
9 yrsof mentorship
Data Mentor
Graduate students · 2015–2024
Mentored Master's and PhD students on statistics and R/Python for journal submissions. Several mentees published with these analyses.
Teaching beyond the classroom
Bite-size lessons on YouTube, plus years of volunteer teaching in ESL, digital literacy and science outreach.
BitSize on YouTube ↗