wtf is rna-syn-bench?

nanda-aditya/rna-syn-bench — explained in plain English

Analysis updated 2026-07-04 · repo last pushed 2026-06-04

★ 0PythonAudience · researcherComplexity · 4/5MaintainedSetup · hard

Mindmap

mindmap
  root((repo))
    What it does
      Tests synthetic gene data
      Checks biological signals
      Measures privacy risk
    Use cases
      Share data safely
      Validate before sharing
      Compare generators
    Cohorts included
      Lung cancer
      Sepsis
      Pediatric IBD
    Methods benchmarked
      Deep learning approach
      Statistical method
    Audience
      Computational biologists
      Data scientists

mindmap root((repo)) What it does Tests synthetic gene data Checks biological signals Measures privacy risk Use cases Share data safely Validate before sharing Compare generators Cohorts included Lung cancer Sepsis Pediatric IBD Methods benchmarked Deep learning approach Statistical method Audience Computational biologists Data scientists

Why would anyone build with this?

REASON 1

Test whether synthetic patient gene data preserves real disease signals before sharing it with outside labs.

REASON 2

Compare different synthetic data generation methods to see which produces the most useful fake datasets.

REASON 3

Train a machine learning model on synthetic data and check if it can accurately classify real patients.

REASON 4

Generate final benchmark charts from pre-computed summary files without accessing raw patient data.

What's in the stack?

PythonRNA-seqMachine LearningDeep Learning

How it stacks up

	nanda-aditya/rna-syn-bench	0xhassaan/nn-from-scratch	3ks/embedoc
Stars	0	0	—
Language	Python	Python	Python
Last pushed	2026-06-04	—	2023-06-08
Maintenance	Maintained	—	Dormant
Setup difficulty	hard	moderate	hard
Complexity	4/5	4/5	1/5
Audience	researcher	developer	developer

Figures from each repo's GitHub metadata at analysis time.

How do you spin it up?

Difficulty · hard Time to first run · 30min

Reproducing full results requires raw patient data locked behind a managed access agreement, charts can be regenerated from included summary files without raw data.

Wtf does this do

When medical researchers want to share patient gene expression data to speed up discoveries, they run into a major privacy hurdle: the data is so detailed that patients could potentially be re-identified. One workaround is to share "synthetic" data, fake patient cohorts generated to mirror the real ones. But how do you know if the fake data is actually good enough to use? This project provides a rigorous scoring system to test synthetic gene expression data on three fronts: whether it preserves real biological signals, whether it's useful for training machine learning models, and whether it truly protects patient privacy. The benchmark evaluates synthetic data across three key areas. First, it checks "differential expression fidelity", meaning, do the fake patients show the same gene-level disease signals as the real patients? Second, it tests machine learning utility by training a predictive model on the synthetic data and seeing if it can accurately classify real, held-out patients. Third, it runs a privacy risk check to ensure the synthetic data generator didn't simply memorize and reproduce real patients, which would defeat the purpose of using fake data. This tool is designed for computational biologists, data scientists, and researchers working with RNA-seq data (a common method for measuring gene activity). For example, a research hospital might want to share a dataset of lung cancer patients with an external lab. Before doing so, they would use this benchmark to evaluate the quality of their synthetic dataset to ensure the external lab could still discover the correct gene mutations and disease patterns without accessing the real, sensitive patient files. The project includes pre-computed results for four specific medical cohorts, including lung cancer, sepsis, and pediatric inflammatory bowel disease. Notably, you can reproduce all the final charts directly from the included summary files without needing access to the raw patient data. Running the full pipeline from scratch requires the raw data, which is locked behind a managed access agreement to protect the original patients. The project benchmarks three different synthetic data generation methods, including a deep learning approach and a simpler statistical method, providing a ready-made framework to compare how different generation techniques perform.

Yoink these prompts

Prompt 1

Using the rna-syn-bench framework, write a script that loads my synthetic RNA-seq dataset and runs all three benchmark evaluations: biological signal fidelity, ML utility, and privacy risk.

Prompt 2

Help me compare three synthetic RNA-seq generation methods using this benchmark's scoring system and produce a summary chart showing which method best preserves disease signals.

Prompt 3

I have synthetic and real RNA-seq data. Walk me through training a classifier on the synthetic data and evaluating it on held-out real patients using this benchmark's ML utility test.

Prompt 4

Using the included pre-computed summary files, generate the final benchmark charts for the lung cancer and sepsis cohorts without needing access to the raw patient data.

Frequently asked questions