NASA Ames — Rohin Sood

Spaceflight biology brings together small datasets, many measured genes, and questions that matter beyond a model score. At NASA Ames, I investigated how predictive models can help identify gene candidates that remain informative across studies and representations.

Two related lines of work

My research included GRIP, a framework for investigating reproducible gene–phenotype relationships, and a separate retinal transcriptomics study. They share a concern with interpretable model evidence, but they are distinct projects with different datasets and publication status.

GRIP: inspect the stability of a ranking

I proposed GRIP to investigate gene relationships across technical representations of RNA-sequencing data. The current framework combines continuous regression, transformation-based environments, invariant model variants, and ensemble gene scoring. Its detailed write-up separates what the implementation measures from stronger causal claims.

Retinal inference: connect expression to injury

For the retinal work, I contributed cross-model feature comparisons, pathway-level interpretation, and comparative results and figures to the study published in npj Microgravity in May 2026. The original analysis notebooks were developed by James Casaletto within the broader collaboration.

A prediction is a starting point

A feature ranking is evidence from a model, not proof of a biological mechanism. I care about making that evidence traceable: which data, which representation, which model, and which assumptions produced the result? The articles below walk through the actual methods and their limitations.