ForetoData / work / oxic-nitrogen-fixation

Choosing which genes to test for oxygen-tolerant nitrogen fixation

Scientific machine learning

A decision system using scientific machine learning to combine biological evidence and rank gene candidates for scarce experimental time and resources.

Case study

Decision to improve

Which undercharacterized genes should researchers test next to understand how some cyanobacteria fix nitrogen in the presence of oxygen?

What shapes the outcome

Nitrogenase is sensitive to oxygen, yet some cyanobacteria coordinate oxygen-producing photosynthesis with nitrogen fixation. Researchers must choose a small set of genes to test from a much larger search space where many functions remain uncertain and few positive examples are established.

Constraints and uncertainty

  • Only a small number of confirmed examples are available to guide a genome-wide search.
  • Useful evidence is spread across gene expression, protein abundance, conservation, promoter structure, and genomic context.
  • The output must help researchers choose experiments, not stop at a prediction score.

How the system works

  1. Defined the experimental decision and assembled reference genes supported by the literature.
  2. Combined nitrogen step-down transcriptomics, quantitative proteomics, promoter features, genomic context, and comparative conservation.
  3. Compared interpretable and nonlinear models, then reviewed the specific evidence supporting high-priority candidates.
  4. Produced a ranked, reviewable set of hypotheses that researchers could use to plan experimental validation.

What this system enables

The work produced a published, reproducible framework for choosing FOX gene candidates to test while keeping computational evidence clearly separated from experimental validation.

What I built

I led the decision framing, data curation, machine-learning methodology, software development, formal analysis, and initial manuscript draft, and contributed to visualization and biological interpretation.