Computational Biologist II · Broad Institute of MIT and Harvard

Computational biology, ML/AI, and scientific systems for drug discovery.

I combine biological domain expertise, machine-learning engineering, and scalable data systems to turn functional-genomics, perturbation, and multi-omics data into target, mechanism-of-action, candidate-prioritization, and experimental decisions.

See selected impact View résumé Contact

100K+ small-molecule inhibitors analyzed 10B+ compounds in virtual screening Top 1% prioritized for follow-up

Selected Work

Biology, ML/AI, and infrastructure working as one system.

Five projects show how I move from biological questions to quantitative methods, reliable systems, and decisions about what to test next.

Functional genomics · Perturbation biology · Broad Institute

Leading large-scale perturbation screening from assay signal to experimental follow-up.

PROSPECT uses large-scale pooled perturbation screens to identify compound-response patterns, prioritize hits, and generate mechanism-of-action hypotheses. I lead computational development for 100K+ small-molecule inhibitors, spanning barcode and replicate QC, dose-response modeling, clustering, hit prioritization, and mechanism-of-action inference. The resulting evidence supports go/no-go decisions and prioritization of the top 1% for follow-up.

Scale: 100K+ compounds
Ownership: Computational lead
Outcome: Top 1% prioritized

Read the research case study →

Virtual screening · MoA inference · Candidate prioritization · Compound triage

Turning virtual-screening predictions into biologically informed compound-selection decisions.

I built the analytical pipeline for mechanism-of-action inference and candidate prioritization, contributed to selection-algorithm design, and evaluate strategies for scaffold generalization and biological relevance. The workflow extends into deterministic medicinal-chemistry and ADMET triage for final scientist review.

Scale: 10B+ compounds
Analysis: MoA + scaffold-aware prioritization
Decision: Human-reviewed compound triage

See methods and final triage →

Scientific AI · Knowledge graphs · Data integration

Connecting biological evidence for reviewable AI-assisted hypothesis generation.

Genomic, metabolic, and pathway evidence is difficult to reason over when it remains fragmented. I build provenance-aware biological knowledge systems that model gene relationships and support LLM-assisted hypothesis generation with traceable evidence. My public MetaMorph work extends the same systems mindset to agentic metadata extraction, validation, and structured scientific data products.

Explore scientific AI and data systems →

Spatial transcriptomics · Graph ML · Disease biology

Learning from spatial and expression structure in breast tissue.

For a collaborative spatial-transcriptomics study, I co-implemented, optimized, and trained a three-layer graph convolutional network combining tissue coordinates with a six-gene extracellular-matrix signature. The model reached AUROC 0.91; the analysis also exposed minority-class limitations and the need for external validation.

Review the model evidence →

Multi-omics · Biomarkers · Translational biology

Integrating transcriptomic and epigenetic evidence across a human cohort.

I integrated adolescent RNA-seq and cord-blood DNA methylation across 250+ pediatric biospecimens to identify heavy-metal exposure signatures connected to immune and neurodevelopmental pathways. I also built automated RNA-seq workflows that improved processing speed by 50% and made the analysis reproducible across the cohort.

Read the research record →

From the Work

Scientific structure you can inspect.

Selected public analyses show how I use visual evidence to test whether computational models preserve meaningful biological structure.

Graph representation connecting spatially adjacent breast-tissue observations — Spatial transcriptomics · Graph MLSpatial neighborhoods connect tissue position with extracellular-matrix expression.

Single-cell latent embedding colored by inferred developmental time — Single-cell · Representation learningLatent structure tests whether developmental time remains recoverable from expression.

Broad Institute descriptions use public or aggregate information and exclude confidential research details.

Translational Differentiator

A computational perspective grounded in translational biology.

My central identity is computational biology; toxicology and preclinical research strengthen how I connect molecular evidence to biological consequence.

Formal training in bioinformatics and toxicology, combined with hands-on IND-enabling research, gives me a working perspective across molecular data, dose-response biology, pharmacology, biomarkers, and the experimental decisions connecting discovery to translation.

See the experience and training →

Training: MS Bioinformatics + MS Toxicology
Preclinical context: IND-enabling PK and toxicology under GLP
Applied lens: Dose · mechanism · biomarkers · data integrity

Technical Expertise

Three capabilities, integrated around scientific decisions.

I understand the biology, develop and evaluate ML/AI methods, and engineer the systems that make those methods reliable at scale.

Computational Biology

Biological questions and decision context.

Functional genomics, CRISPRi, and perturbation screening
Target discovery, mechanism of action, and biomarkers
Multi-omics, spatial transcriptomics, and translational biology

ML/AI Engineering

Methods built and evaluated for scientific use.

Representation learning, graph neural networks, and virtual screening
Model benchmarking and selection-algorithm evaluation
Knowledge graphs, RAG/LLM workflows, and agentic systems

Scientific Data Systems

Infrastructure that carries analysis into practice.

Python, R, SQL, APIs, and structured scientific data products
Workflow orchestration, containers, testing, and provenance
HPC and cloud computing across AWS and Azure

Integrated outcome: scalable evidence for target and mechanism interpretation, candidate prioritization, biological hypotheses, model selection, and follow-up experiments.

Additional Experience & Deeper Record

Scientific systems and evidence beyond the flagship projects.

Additional work demonstrates production-minded genomics infrastructure, while the deeper pages document research methods, public software, publications, and training.

Scientific informatics · APIs · Parallel computing

Reducing microbial-genomics turnaround with scalable data infrastructure.

I optimized API-driven retrieval, probabilistic filtering, and parallel processing across 15M+ sequences and 11+ taxa, reducing diagnostic turnaround by more than two days. The work demonstrates scientific data engineering tied directly to a biological operating need.

See the full experience →

Research · Public software · Full experience

Follow the evidence beyond the homepage.

Research details expose methods and limitations; public repositories show how I structure scientific systems; the full experience record connects that work to training and experimental context.

Connect

Building computational biology, scientific ML, or research data systems?

I am interested in roles where biological depth, ML/AI engineering, and scalable scientific infrastructure come together to inform experimental and therapeutic decisions.

Start a conversation Open résumé →