MS Bioinformatics candidate at Northeastern University— I build reproducible computational pipelines for transcriptomics, single-cell, and mRNA biology.
I work at the intersection of RNA biology and scientific computing: turning raw sequencing and structural data into reproducible, interpretable results. I write pipelines and analysis code (Nextflow DSL2, Python, R/Bioconductor) and run them reproducibly on HPC with containers and SLURM. Focus on differential expression, cell-type annotation, and model-vs-baseline benchmarking.
- Transcriptomics & RNA-seq — differential expression, quantification, functional enrichment
- Single-cell genomics — cell-type annotation, foundation models vs. classical baselines
- mRNA structure & function — secondary-structure analysis, sequence-to-function modeling
- Reproducible pipelines — Nextflow (DSL2), Docker/Apptainer, SLURM/HPC, GPU workflows
scRNA-seq foundation models vs. a simple baseline: do they actually win?
Benchmarked a fine-tuned Geneformer foundation model against a classical PCA + logistic-regression baseline for cell-type annotation on human yolk-sac scRNA-seq data. The simple baseline matched or beat the foundation model (macro-F1 0.93 vs. 0.85), with the gap traced to rare, biologically similar progenitor populations — independently reproducing recent published benchmarks. Built as a reproducible HPC pipeline: stratified data handling, GPU fine-tuning via SLURM (H200/V100), integrity-checked evaluation, UMAP + confusion-matrix visualizations, and a containerized (Apptainer) environment for one-command reproduction.
Python · PyTorch · Geneformer · scanpy · SLURM · Apptainer
Epileptic seizure prediction — deep neural network on EEG
A deep neural network that detects epileptic seizures from raw EEG signals, benchmarked against a K-Nearest-Neighbors baseline to test whether the added model complexity actually earns its keep. Full pipeline in notebooks — EEG preprocessing, augmentation, training, and evaluation built to be opened and re-run, not just read about.
Python · TensorFlow · scikit-learn · EEG · Jupyter
nf-rnaseq-mettl3 — reproducible RNA-seq pipeline (Nextflow DSL2)
Custom Nextflow DSL2 pipeline (Salmon → DESeq2 → GO/GSEA), containerized and deployed on HPC, analyzing the transcriptomic response to METTL3 (m6A writer) knockdown. End-to-end reproducible from raw FASTQ to interpreted biology, with a test profile, SLURM execution, and a single self-contained container image.
Nextflow · R/Bioconductor · Salmon · Docker/Apptainer · SLURM
Languages
Genomics & ML
Infrastructure
