Sangyu Xu 徐桑榆
  • Home
  • Neuroscience
  • OSS Tools
  • Communications
  • Demos
    • DABEST 2.0
    • Gravitas
    • Whorlmap
    • Rugprint
    • MODA
    • ESPLoCo
  • Blog
  • Art & Music
  • CV
  • xusangyu.com

Open Source Tools

Open Source Tools

Measurement frameworks for complex, ambiguous systems.

DABEST estimation plot: raw data above, mean differences with bootstrap distributions below
Nature Methods 2026

DABEST 2.0

Bootstrap-coupled effect size estimation for multi-group comparisons. Replaces significance testing with estimation graphics: mean differences, confidence intervals, and the full sampling distribution of your effect. The delta-delta plot handles 2×2 factorial designs and beyond.

→ Nature Methods   Read more   Demo ↗   Colab ↗   GitHub ↗
DESTRA phenovectors: manipulations clustered by their full behavioral effect-size profile
bioRxiv 2026

DESTRA

Delta ethomic state-transition recapitulation assessment. Bootstrap effect size profiles across all measured metrics, anchored to a ground-truth reference condition. Asks not whether individual metrics differ, but whether the full profile matches the target state.

→ bioRxiv   Read more   GitHub ↗
Gravitas citation map: a seed paper ringed by teal citing papers and orange cited papers, with two expanded satellite clusters

Gravitas

A citation map where every edge carries the sentence behind it. Drop in a DOI, PubMed ID, or title and get the paper's citation neighbourhood; click any edge to read the verbatim sentence the citing paper wrote about the cited one. Claims are extracted by string slicing, never generated, and an edge with nothing retrievable says so instead of inventing a summary.

→ Try it live   Read more   GitHub ↗
Whorlmap grid: each heatmap cell drawn as a spiral encoding its bootstrap distribution

Whorlmap

A compact heatmap design for high-dimensional comparisons where each cell encodes a full bootstrap/effect-size distribution as a tiny spiral, rather than collapsing uncertainty into a single color. Lets readers scan many treatment–outcome contrasts at once while still seeing direction, magnitude, and uncertainty inside each cell.

Demo ↗   GitHub ↗
Rugprint: linked scatterplot projections with edge rugs, one observation tracked across panels

Rugprint

A compact multivariate fingerprint: selected 2D scatterplot projections arranged as a sparse map, with edge rugs — a Tufte idea — showing how the same observations collapse onto one-dimensional axes.

Demo ↗   GitHub ↗
 

Built with Quarto