Knowledge Graph Reveals Cross-Paper Insights in Gut-Brain Axis Research

Building a 42-paper knowledge graph of gut-microbiome-mental-health research in RhizomeDB has surfaced insights no individual paper states:

The SCFA Paradox in Parkinson's Disease

Short-chain fatty acids (SCFAs) are depleted in Parkinson's patients (Unger 2016, Keshavarzian 2015) — suggesting they're protective. But Sampson 2016 showed SCFAs promote alpha-synuclein aggregation and neuroinflammation in germ-free mice. This contradiction only becomes visible when you hold both findings simultaneously in a graph.

Three Conditions, One Causal Pattern

Fecal microbiota transplant from affected humans into germ-free mice reproduces behavioral phenotypes across three independent conditions: Depression (Kelly 2016, Zheng 2016), Parkinson's (Sampson 2016), and ADHD (Tengeler 2020). This three-condition convergence is the strongest argument that the gut-brain axis is mechanistically real, not merely correlational.

Neuroinflammation is the #1 Convergence Hub

With a convergence score of 144 (8 papers × 3 bacteria × 6 conditions), neuroinflammation is the most connected mechanism in the graph. It links depression, anxiety, schizophrenia, bipolar, Parkinson's, and ADHD through shared bacterial pathways.

The Ethnic Invariance Finding

Bos 2022's HELIUS cohort (N=3,211 across 6 ethnic groups including Ghanaian, Moroccan, and South-Asian) showed that microbiota-depression associations are invariant across ethnicities — suggesting the findings from our predominantly Western-population graph are not merely dietary artifacts.

What the Graph Reveals That Papers Don't

• 4 contradictions where different studies find opposite directions for the same bacterium-condition pair
• 245 novel connections between entities that never appear together in any single paper
• Depression and Anxiety share 39% of bacterial associations (Jaccard similarity) while Parkinson's is isolated from both
• Diet-driven microbiome modulation is the highest-leverage mechanism (convergence score 81) but studied by only 3 papers

Built using RhizomeDB's delta/HyperView model with 1,933 deltas, 19 composable query functions, and 471 tests. The knowledge graph treats each claim as a first-class entity with full provenance (paper, authors, institution, country, study type, direction annotation).