# Study shows multi-sample network inference robustness varies with prior knowledge and data

Digest AI · Research · published 2026-10-19T00:00:00Z

Canonical: https://digestai.news/story/study-shows-multi-sample-network-inference-robustness-varies-with-prio

## Summary

Researchers reproduced and extended the CORNETO framework for multi-sample constrained network inference, testing it on a dynamic haematopoiesis dataset and a stable adult brain cortex dataset. They found that joint inference generally improves robustness and reproducibility, but gains in predictive performance depend on the completeness of the prior‑knowledge network (PKN) and the amount of sample data available.

Key experiments showed that multi‑sample sFBA reduced network size by ~32% at 10% of optimal biomass and ~6% at 90%, while the coefficient of variation rose from 2.9% to 8.9% near‑optimal growth. Bootstrap aggregation added 8–11 reactions (≈8–11% expansion) and raised edge overlap to ~89% in multi‑sample settings. Selective re‑expansion of the top 100 informative genes boosted KPNN classification accuracy from 0.576 to 0.906 and macro F1 from 0.424 to 0.848. Prior‑network completeness strongly affected results: at 20% completeness no active reactions were recovered, whereas 80–100% completeness enabled higher precision and recall across regularization strengths. Increasing jointly modelled conditions from 2 to 16 improved Jaccard similarity of inferred reaction sets, indicating greater stability.

The authors conclude that multi‑sample constrained inference is most beneficial when samples share substantial regulatory structure and when the PKN is sufficiently complete, providing practical guidance for applying CORNETO across diverse biological contexts.

## Key points

- Multi‑sample inference cut sFBA network size by ~32% at low biomass and ~6% at high biomass, with variability rising from 2.9% to 8.9%
- Bootstrap aggregation recovered 8–11 additional reactions (~8–11% expansion) and raised edge overlap to ~89% in multi‑sample runs
- Selective re‑expansion raised KPNN accuracy from 0.576 to 0.906 and macro F1 from 0.424 to 0.848

## Why it matters

Understanding when multi‑sample network inference improves robustness helps researchers choose appropriate strategies for omics studies, influencing reproducibility and biological insight.

## Sources

1. [Reusability report: multi-sample omics network inference operates in distinct robustness regimes defined by prior knowledge and data availability](https://nature.com/articles/s42256-026-01311-y) (Nature Machine Learning, 2026-10-19, primary source)

## Cite

Digest AI, "Study shows multi-sample network inference robustness varies with prior knowledge and data", 19 October 2026, https://digestai.news/story/study-shows-multi-sample-network-inference-robustness-varies-with-prio

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/study-shows-multi-sample-network-inference-robustness-varies-with-prio.json