← kz. / notebook

VOL. I · GENE REGULATORY NETWORKS

Plates from
the lab.

Which genes switch which others on and off; inferred from data, then checked against the evidence. Four studies, drawn and annotated.

Cheng Lab · University of Missouri · since Aug 2025

Electron micrograph of cells with labelled nucleus, nucleolus and organelles
THE SUBJECTELECTRON MICROGRAPH
PLATE I ACCEPTED · NeurIPS 2026

EviGRN: multimodal graph reasoning for gene regulatory network inference

A. Hegde, K. Zook, Y. Wang, J. Cheng. NeurIPS 2026.

MY PART: the automated literature-validation agent, and the multi-omic validation step that combines chromatin accessibility with transcriptomics.

code: EviGRN ↗
expressionaccessibilitymotifTF-centred graphliterature
fig. a: evidence in, a graph out, every edge checked against papers
PLATE II IN REVIEW · npj SYS. BIO.

GRNPred: a multimodal graph transformer with masked gene-expression pretraining

T. Nguyen, A. Hegde, K. Zook, J. Cheng. npj Systems Biology and Applications, in review.

MY PART: co-first author. Ran the full-model, ablation and supplementary experiments for revision, fixed the training and evaluation code, rewrote parts of the manuscript.

code: GRNPred ↗
expressionGO termsgene textmotif priorsi. masked pretrainii. edge finetuneTF → target
fig. b: learn expression first, then learn the edges
PLATE III IN PREPARATION

Dynamic gene regulatory network inference (working title)

A. Hegde, K. Zook, J. Cheng.

THE IDEA: networks that change over time, rather than one fixed wiring diagram.

t₁t₂t₃
fig. c: same cell, different wiring at each step
PLATE IV FIRST AUTHOR · IN PREP

A benchmark for evaluating graph-based gene regulatory network inference

K. Zook, A. Hegde, J. Cheng. Planned for submission, December 2026.

THE IDEA: run every method through one shared evaluation, so the results can actually be compared.

shared evaluationmethodscomparable
fig. d: one ruler for every method