System Graph

Modeling the world as one interconnected system

The System Graph leverages state of the art LLMs, advanced graph technology, complexity science, and causal inference techniques to generate a systems model of the world — unifying scales and domains to connect siloed knowledge.

Features of the System Graph

How it works

Using AI, the System Graph extracts, grounds, and links findings from verified sources like peer-reviewed studies and expert-curated databases.

Release notes

Integrating Retraction Watch Database with System Graph

Study
Type-Specific Persistence of Human
Papillomavirus DNA before the
Development of Invasive Cervical Cancer
cited by
473
journal
The New
England Journal
of Medicine
publish date
11/25/1999
variable 1
Positive test for human papillomavirus DNA
variable 2
Development of invasive cervical cancer
Population
118 women in Sweden who participated in a population-based screening program for cancer of the cervix from 1969 to 1995.
Statistics
Odds Ratio 16.400
p-value = 0.001
95% CI (4.400 to 75.100

Using AI, the System Graph extracts, grounds, and links findings from verified sources like peer-reviewed studies and expert-curated databases.

This generates millions of rich, quantitative relationships that cut across sources and fields.

Those quantitative relationships are joined with mechanistic relationships to link scales of knowledge.

From knowledge graph to System Graph

FromTo
ArchitectureSemantic web and knowledge graphsSystem Graph
ArchitectureSemantic web and knowledge graphsSystem Graph
From
To
Architecture
Semantic web and knowledge graphs
System Graph
Primary node
Concepts
Variables linked to data
Primary edge
Semantic relationship
Statistical meta-analysis
Applications
Object retrieval descriptive
System retrieval
Contextual, diagnostic, and predictive

LLM-based Foundation Models vs. System Graph

FromTo
ArchitectureSemantic web and knowledge graphsSystem Graph
ArchitectureSemantic web and knowledge graphsSystem Graph
LLM-Based Foundation Models
System Graph
Data sources
Range of unknown sources, including unverified ones
Facts extracted only from trusted, verified sources
Data structure
Based on tokens and their embeddings
Composed of normalized entities and quantitative relationships
Reasoning
Non-deterministic
Systemic (multi-hop upstream causes and downstream effects)
Knowledge cut-off
Updated periodically
Updated daily
Interpretability
Limited
Full knowledge provenance
Risk of hallucination
High
Low

Currently recognized association types

Statistical

Adjusted hazard ratio

Adjusted odds ratio

Coefficient of determination

Elasticity

Gini coefficient

Hazard ratio

Incident rate ratio

Mean difference standardized

Mean difference unstandardized

Odds ratio

Pearson correlation coefficient

Prevalence ratio

Regression coefficient

Relative risk ratio

Risk difference

Spearman correlation coefficient

T-test

Mechanistic

Acetylation

Activation

Complex

Conversion

Deacetylation

Decrease amount

Defarnesylation

Degeranylgeranylation

Deglycosylation

Dehydroxylation

Demethylation

Demyristoylation

Depalmitoylation

Dephosphorylation

Deribosylation

Desumoylation

Deubiquitination

Farnesylation

Geranylgeranylation

Glycosylation

Hydroxylation

Increase amount

Inhibition

Methylation

Modification

Myristoylation

Palmitoylation

Phosphorylation

Ribosylation

Sumoylation

Translocation

Ubiquitination

Ever-expanding corpus 
of verified source material

Case study

Integrating the EMBL GWAS Database into the System Graph

Integrated sources

PubMed
GWAS Catalog
Comparative Toxicogenomics Database
PubMed Central
BioGrid
OpenAlex
1
findings

224 added today

No change

1
concepts

224 added today

No change

1
relationships

3 added today

No change

Surfaced by the System Graph and extracted from verified sources including peer-reviewed studies and expert-curated databases

As of 8:01am EST on 8/19/24

Ever-expanding corpus of verified source material

Integrated sources

PubMed

PubMed Central

GWAS Catalog

BioGrid

Comparative Toxicogenomics Database

OpenAlex

Case study

Integrating the EMBL GWAS Database into the System Graph

Read  more

An open and interoperable public resource

Interpretable and transparent with complete knowledge provenance

Primary nodes are variables that are interoperable with established ontologies

Accessible and queryable via GraphQL, REST APIs, or natural language

Tech

Classifying Study Type Using Deep Learning

Why haven’t we seen it all connected yet?