Research, insights, and ideas
from the Bystro team.

Deep dives into genomics, the science behind THiNK, and how personalized health is evolving.

Research2026-09-038 min read

Complex Figure Creation via Natural Language, an AI Benchmarking Study

Working from plain language instructions, an agentic AI assistant can produce the same visualizations from complex clinical trial data as a skilled human analyst

Read article

Complex Figure Creation via Natural Language, an AI Benchmarking Study

Research2026-08-277 min read

Writing a Peer-Reviewed Machine Learning Paper in 30 Days

I gave myself one month to turn a real diagnostic problem into a conference paper. Bystro helped shape the method, implement baselines, search the literature, write the manuscript, and prepare for peer review. The paper was accepted at MLHC 2026.

Read article

Writing a Peer-Reviewed Machine Learning Paper in 30 Days

Research2026-08-205 min read

When an AI Analyst Meets a 10000+ Patient Clinical Trial (and Gets It Right)

Working from plain language instructions, an agentic AI assistant can produce the same results from complex clinical trial data as a skilled human analyst

Read article

When an AI Analyst Meets a 10000+ Patient Clinical Trial (and Gets It Right)

Research2026-08-075 min read

Using Bystro to Assist with Rebuttals at Machine Learning for Healthcare

Reviewer scores almost never move after a rebuttal, but ours did. Here's how Bystro helped us turn our harshest MLHC reviewer's 2 into a 3.

Read article

Using Bystro to Assist with Rebuttals at Machine Learning for Healthcare

Research2026-07-077 min read

Ask Bystro: Can We Catch Critical Typos in Published Manuscripts?

We asked Bystro to audit Ahmad (2022), a statistical paper on high-dimensional covariance proportionality. It found a wrong coefficient in a trace estimator and a missing factor of 2 in the null calibration, then reproduced the failure with spot simulations.

Read article

Ask Bystro: Can We Catch Critical Typos in Published Manuscripts?

Research2026-06-227 min read

Detecting Parent-of-Origin Effects Without Family Data: Introducing POISE

Parent-of-origin effects shape growth, metabolism, and neurodevelopment, but detecting them at scale has required scarce family data. POISE, our new spectral method, recovers these effects from standard GWAS cohorts, outperforms existing tests, and identifies 134 novel loci in the UK Biobank.

Read article

Detecting Parent-of-Origin Effects Without Family Data: Introducing POISE

Research2026-06-117 min read

Discovering Interpretable Omic Networks with gPCR.

Standard SVAES can predict outcome, but not the true network behind it. A new method called gPCR does both, and could transform how scientists design experiments.

Read article