Research/Earlier Research
Generative models
for microbiome data.
Research experience in Professor Li Ma’s group at Duke University.
- Group
- Professor Li Ma, Department of Statistical Science, Duke
- Period
- May – Dec 2024
- Data
- American Gut dataset
- Role
- Collaborative research
Schematic · not dataWhat does a generated sample preserve?
A generative model can produce samples that look plausible. The harder question is whether they keep what matters for the analyses people run on real microbiome data.
Models, compared at scale.
I studied different generative models and neural networks on their ability to generate microbiome data, with an emphasis on generating large datasets, and ran the work on large computing clusters using the American Gut dataset.
By downstream analyses.
The quality of generated data was evaluated with bioinformatics analyses, using R packages for microbiome data.
What this page does not show.
- No results are shown here.
- The figure is a schematic of observed and generated compositions, not data.