Emil Dmitruk

MesoSCOUT & topological data-analysis tooling

Open-source · Julia · first-author preprint (2025)

The problem

Large brain networks (and many other real-world networks) are usually compared by counting connections. But raw counts are dominated by a few very strong links and by whatever threshold you happened to pick — so genuine mesoscale structure, the mid-level organization between individual edges and the whole graph, gets washed out.

MesoSCOUT

MesoSCOUT reveals that mesoscale structure from the relative strength of connections rather than absolute counts, using a rank-based topological design (persistent homology on weighted networks). Because it is rank-based it is robust to the acquisition differences that normally make datasets impossible to compare directly — I demonstrated this by finding consistent structure across two independent cohorts (COBRE, N = 88; HCP).

Connectivity structure compared across independent cohorts with MesoSCOUT

The same mesoscale structure recovered across independent datasets — see the visualization showcase for how these are built and read.

Engineering

The work is backed by reproducible analysis pipelines and a set of open-source Julia packages I authored for topological data analysis:

These pipelines emphasize the engineering rigour I bring to research code: pure functions, memoization and algorithmic caching (a 10–15× speed-up over the original implementation), and reproducibility.

MesoSCOUT grew out of several years of first-authored conference work applying topology to brain connectivity and perception (OCNS 2020–2023). Those results are collected on the Publications page.

CC BY-SA 4.0 Emil Dmitruk. Last modified: July 22, 2026. Website built with Franklin.jl and the Julia programming language.