Chapter 4: From simple networks to multisource integration
The introduction frames the move from a single ontology to multisource knowledge graphs and explains why traditional integration, rather than LLMs, leads their construction.
After this chapter you will be able to
- Shift to multisource graphs. Explain how this chapter differs from the single-ontology approach in chapter 3: it uses multiple structured sources in graph-friendly formats, allowing attention to graph modeling, integration, and analysis.
- Limited LLM construction role. Explain why traditional data integration is the primary approach to constructing these KGs and LLMs have a complementary, auxiliary role.
- Biomedical graph question answering. Explain that biomedical tasks can be approached by organizing existing knowledge as a graph and checking whether it contains the information needed to answer a complex question.
- Multi-omic biological layers. Explain what multi-omic analysis combines and distinguish the genome, transcriptome, and proteome in the sequence described here.
- Heterogeneous source integration. Describe how the cited gene–symptom KG brought together heterogeneous data sources and made their disease terms consistent.
- Disease pathway network rationale. Explain why protein interaction and protein–disease networks are useful for investigating disease pathways rather than examining disease proteins separately.
- Disease pathway graph structure. Explain how known disease–protein associations and protein–protein interactions form the starting graph for discovering potential disease pathways.
- Multisource graph assembly. Describe what each imported data source contributes to the KG and why gene information is added.
- PPI network connectivity. Explain what weakly connected components reveal about the PPI network and interpret the reported result.
- Louvain protein communities. Distinguish Louvain communities from WCC components and describe what the Louvain results show.
- Disease pathway subgraphs. Describe how a disease pathway is represented within the PPI network and how its boundary relates it to the rest of the network.
- Internal pathway connectivity measures. Explain what the largest pathway component fraction and pathway density reveal about connections among disease-associated proteins.
- Pathway conductance interpretation. Explain how conductance characterizes a disease pathway’s separation from the rest of the PPI network.
- Pathways versus Louvain clusters. Contrast the reported connectivity patterns of disease pathways with those of Louvain clusters.
- Drug repurposing with KGs. Explain why drug analysis and repurposing are valuable applications for KGs, and identify a limitation of predicting associations only from drugs and diseases already in a graph.
- Hetionet as integrated knowledge. Describe what Hetionet brings together and why its schema matters for exploring the graph.
- Metapaths versus actual paths. Distinguish a metapath’s type-level description from an actual path between nodes in Hetionet.
- Path counts and degree weighting. Explain how PC and DWPC measure paths matching a metapath, including how DWPC weights and combines individual paths.
- DWPC-based GO process ranking. Explain how the analysis identifies GO processes linked to celiac disease and why ranking by DWPC gives a different result from ranking by PC.
- Refined tissue-specific interactomics. Describe how the second query refines the disease-to-process analysis and what changed in its results.
- LLM interpretation of pathway rankings. Explain why DWPC-based pathway rankings need contextual interpretation and how an LLM can turn them into a biological narrative.
- Clinical data integration. Explain why clinical KGs bring together omics data, EHRs, ontologies, and other sources to support the long-term goal of precision medicine.
- Privacy-conscious clinical graphs. Describe the two privacy-conscious approaches to clinical KGs presented here.
- CKG clinical investigation. Describe what the Clinical Knowledge Graph integrates and how the section’s protein–disease query provides a starting point for investigation.
- Clinical KG interpretation. Explain how an LLM can help interpret complex clinical KG relationships and place a finding, such as the association between ACTC1 and cardiomyopathies, in a clinical context.
- Human clinical oversight. Justify why LLM-generated clinical insights require interpretation by physicians and researchers rather than being used directly.
What it builds on
- Ontology ingestion and processing with neosemantics (chapter 3) · The earlier work of building a KG from an ontology.
- Building a knowledge graph (chapter 3) · The reader has previously encountered building a KG by ingesting data.
- A KG approach to IASs (chapter 2) · the role of knowledge graphs in intelligent application systems
Section by section
Introduction
The introduction frames the move from a single ontology to multisource knowledge graphs and explains why traditional integration, rather than LLMs, leads their construction.
- Shift to multisource graphs: Explain how this chapter differs from the single-ontology approach in chapter 3: it uses multiple structured sources in graph-friendly formats, allowing attention to graph modeling, integration, and analysis.
Why it matters: This shift explains the chapter’s emphasis on combining and analyzing sources rather than working with one ontology. - Limited LLM construction role: Explain why traditional data integration is the primary approach to constructing these KGs and LLMs have a complementary, auxiliary role.
Why it matters: It sets appropriate expectations for where LLMs fit when the source data is already structured.
Key terms: knowledge graphs (KGs), ontology, structured data sources, graph-friendly formats, graph modeling decisions, integration strategies, LLMs, CSV files, relational databases, APIs
4.1 Biomedical knowledge graphs and applications
The section introduces biomedical knowledge graphs as a way to organize existing knowledge for questions across applications with different goals and data needs.
- Biomedical graph question answering: Explain that biomedical tasks can be approached by organizing existing knowledge as a graph and checking whether it contains the information needed to answer a complex question.
Why it matters: This connects the value of a biomedical KG to the question it can actually answer. - Application goals and data sources: Describe the range of biomedical KG applications in the section and explain that different applications have different business goals and use different data sources.
Why it matters: The intended application helps determine which biomedical knowledge is relevant to the graph.
Key terms: biomedical knowledge, graphs, complex questions, business goals, data sources, biomedical applications
4.2 Multi-omic applications of KGs
This section defines multi-omic data and shows why biomedical knowledge graphs bring biological layers and disparate data sources together to investigate disease.
- Multi-omic biological layers: Explain what multi-omic analysis combines and distinguish the genome, transcriptome, and proteome in the sequence described here.
Why it matters: These layers explain what kinds of biological information a multi-omic KG can connect. - Heterogeneous source integration: Describe how the cited gene–symptom KG brought together heterogeneous data sources and made their disease terms consistent.
Why it matters: Combining sources is more useful when their differing identifiers can refer to the same disease consistently. - Disease pathway network rationale: Explain why protein interaction and protein–disease networks are useful for investigating disease pathways rather than examining disease proteins separately.
Why it matters: The relationships among proteins can reveal disease-relevant patterns that isolated protein descriptions miss.
Key terms: multi-omic, genome, transcriptome, proteome, transcription, heterogeneous data sources, disease identifiers, Unified Medical Language System, UMLS codes, candidate genes
4.2.1 Creating a KG from the PPI and protein-disease networks
The section describes how to assemble a disease-pathway knowledge graph from protein interactions, disease associations, disease categories, and gene information, then check the imported data.
- Disease pathway graph structure: Explain how known disease–protein associations and protein–protein interactions form the starting graph for discovering potential disease pathways.
Why it matters: This graph structure explains what the imported data represents and what pathway discovery is intended to find. - Multisource graph assembly: Describe what each imported data source contributes to the KG and why gene information is added.
Why it matters: Each source adds a different kind of connection or context needed to explore the graph. - Import constraints and checks: Explain why node key constraints and count checks accompany these imports, including why the protein count may change after disease associations are added.
Why it matters: These checks help distinguish the original interaction network from proteins added through later associations.
Key terms: disease pathways, PPI network, disease–protein association network, monopartite graph, bipartite graph, SNAP, protein–disease associations, disease categories, Disease Ontology, UMLS codes
4.2.2 High-level analysis of the resulting KGs
The section uses WCC and Louvain to inspect the PPI knowledge graph, showing the difference between overall connectivity and densely connected communities.
- PPI network connectivity: Explain what weakly connected components reveal about the PPI network and interpret the reported result.
Why it matters: This establishes that most proteins belong to one connected network before looking for finer-grained groups. - Louvain protein communities: Distinguish Louvain communities from WCC components and describe what the Louvain results show.
Why it matters: A mostly connected network can still contain meaningful internal groupings. - Generic analysis limits: Explain the limitation of these graph algorithms for domain-specific analysis.
Why it matters: Structural clusters alone do not provide an analysis tailored to the biomedical question.
Key terms: weakly connected component (WCC), PPI network, connected components, Louvain modularity algorithm, community, modularity score, generic, node, relationship
4.2.3 Domain-specific analysis of the PPI and disease KG
Defines disease pathways as protein subnetworks, measures their connectivity, and compares their structure with Louvain clusters.
- Disease pathway subgraphs: Describe how a disease pathway is represented within the PPI network and how its boundary relates it to the rest of the network.
Why it matters: The subgraph definition determines what the pathway measures are measuring. - Internal pathway connectivity measures: Explain what the largest pathway component fraction and pathway density reveal about connections among disease-associated proteins.
Why it matters: These measures distinguish a pathway’s fragmentation from how densely its proteins are linked. - Pathway conductance interpretation: Explain how conductance characterizes a disease pathway’s separation from the rest of the PPI network.
Why it matters: Conductance captures external connectivity that the two internal measures do not. - Pathways versus Louvain clusters: Contrast the reported connectivity patterns of disease pathways with those of Louvain clusters.
Why it matters: The comparison shows that disease association and graph-based clustering pick out structurally different subnetworks.
Key terms: PPI network, disease pathway, undirected subgraph, monopartite subnetwork, pathway boundary, largest pathway component, density, connected component, conductance, disease pathways
4.3 Pharmaceutical applications of KGs
Introduces Hetionet as a multisource pharmaceutical knowledge graph and explains how metapaths, path counts, and degree-weighted path counts support its analysis.
- Drug repurposing with KGs: Explain why drug analysis and repurposing are valuable applications for KGs, and identify a limitation of predicting associations only from drugs and diseases already in a graph.
Why it matters: This explains the pharmaceutical problem that motivates a richer knowledge graph. - Hetionet as integrated knowledge: Describe what Hetionet brings together and why its schema matters for exploring the graph.
Why it matters: The graph’s multiple entity and relationship types make its schema important to interpreting and querying it. - Metapaths versus actual paths: Distinguish a metapath’s type-level description from an actual path between nodes in Hetionet.
Why it matters: The distinction lets you interpret the path metrics as measurements of instances matching a specified pattern. - Path counts and degree weighting: Explain how PC and DWPC measure paths matching a metapath, including how DWPC weights and combines individual paths.
Why it matters: Degree weighting keeps numerous paths through heavily connected nodes from dominating a relevance measure simply because those nodes have many connections.
Key terms: drug repurposing, drug–disease associations, molecular targets, Hetionet, heterogeneous network, schema, node types, relationship types, metapath, real paths
4.3.1 Deep analysis of the Hetionet knowledge graph
This section uses Cypher queries and DWPC to identify disease-relevant GO processes in Hetionet, refine the analysis with biological evidence, and inspect the paths behind a result.
- DWPC-based GO process ranking: Explain how the analysis identifies GO processes linked to celiac disease and why ranking by DWPC gives a different result from ranking by PC.
Why it matters: A high path count alone can make a widely connected process look more disease-specific than it is. - Refined tissue-specific interactomics: Describe how the second query refines the disease-to-process analysis and what changed in its results.
Why it matters: Changing the evidence and paths considered can expose disease-specific biology that the initial ranking did not emphasize. - Paths behind a DWPC: Explain how the analysis follows up on an interesting process result.
Why it matters: The score points to a process, while its underlying paths let a reader explore the stored connections.
Key terms: DWPC, PC, GO process, T cell costimulation, GWASs, protein interaction, metapath, upregulated, interactome, paths behind the DWPC
4.3.2 LLM-assisted interpretation of pathway analysis results
This section shows how an LLM can add biological context to quantitative knowledge-graph pathway rankings for celiac disease.
- LLM interpretation of pathway rankings: Explain why DWPC-based pathway rankings need contextual interpretation and how an LLM can turn them into a biological narrative.
Why it matters: It distinguishes what the graph analysis measures from the context an LLM can help supply. - Celiac pathway interpretation examples: Describe how the example LLM output relates highly ranked immune pathways to celiac disease, including what it suggests about costimulation and tolerance induction.
Why it matters: These examples make concrete the difference between reporting a pathway score and interpreting its possible disease significance.
Key terms: DWPC-based queries, pathway analysis, LLMs, contextual interpretation, T cell costimulation, tolerance induction, oral tolerance, celiac disease, DWPC score
4.4 Clinical applications of KGs
The section explains how clinical knowledge graphs combine biomedical and patient information, address privacy concerns, and support clinical investigation.
- Clinical data integration: Explain why clinical KGs bring together omics data, EHRs, ontologies, and other sources to support the long-term goal of precision medicine.
Why it matters: It explains why a clinical KG needs more than either patient records or molecular data alone. - Patient journey mapping: Distinguish patient journey mapping from a clinical pathway as described in the section.
Why it matters: The distinction separates an individual’s experience of care from a standard care plan. - Privacy-conscious clinical graphs: Describe the two privacy-conscious approaches to clinical KGs presented here.
Why it matters: Clinical graphs must make useful connections without unnecessarily exposing sensitive patient records. - CKG clinical investigation: Describe what the Clinical Knowledge Graph integrates and how the section’s protein–disease query provides a starting point for investigation.
Why it matters: It shows the kind of investigable association an integrated clinical KG can surface.
Key terms: precision medicine, omics data, electronic health records (EHRs), clinical KGs, patient journey mapping, patient experience mapping, clinical pathways, statistical relationships, related weights, deidentified
4.4.1 LLM-guided clinical decision support analysis
This section explains how an LLM can help turn clinical knowledge graph findings into interpretable decision support while leaving clinical judgment to human experts.
- Clinical KG interpretation: Explain how an LLM can help interpret complex clinical KG relationships and place a finding, such as the association between ACTC1 and cardiomyopathies, in a clinical context.
Why it matters: A KG finding becomes more useful when clinicians and researchers can interpret its possible relevance to care or research. - Human clinical oversight: Justify why LLM-generated clinical insights require interpretation by physicians and researchers rather than being used directly.
Why it matters: Interpretable output can aid clinical work without being mistaken for an independently validated decision.
Key terms: clinical KGs, CKG analysis, ACTC1, cardiomyopathies, clinical decision support, physicians and researchers, interpretable insights, IASs
4.5 Summary
Concepts explained in this chapter
- disease pathwayA disease pathway is a subgraph of a protein-protein interaction network containing proteins and interactions associated with a particular disease.
- degree-weighted path countA degree-weighted path count is a path-based graph metric that weights paths according to the degrees of the entities they traverse.
- monopartite graphA monopartite subnetwork is a network containing one type of node, with connections only between nodes of that type.
- protein-protein interaction networkA protein-protein interaction network is a graph whose nodes represent proteins and whose edges represent experimentally observed physical or functional interactions between them.
- knowledge graphA knowledge graph is an evolving graph data structure of typed entities, attributes, and meaningful named relationships that integrates structured and unstructured domain data.
- HetionetHetionet is a biomedical knowledge graph integrating entities and relationships involving diseases, genes, drugs, and other biological concepts.
- collaborative filteringCollaborative filtering is a recommendation technique that infers preferences or associations from patterns across users, items, or entities.
- damping exponentA damping exponent is a nonnegative exponent that controls how graph degree values reduce the contribution of a path.
- drug repurposingDrug repurposing is the investigation of an existing drug for a new therapeutic use.
- metapathA metapath is a sequence of node classes and relationship classes describing possible paths between specified source and destination types.
- path countPath count is the number of paths matching a specified metapath between defined source and target nodes.
- path-degree productPath-degree product is the product of degree values raised to a damping exponent for all edge incidences along a path.
- graph conductanceGraph conductance is a measure of the connectivity between a subgraph and the remainder of its graph relative to the connections incident to the subgraph.
- graph densityGraph density is the fraction of possible edges that are present between nodes in a graph or subgraph.
- precision medicinePersonalized medicine is an approach to healthcare that tailors prevention, diagnosis, or treatment to individual patient characteristics.
- data integrationThe process of combining data from multiple sources into a coherent and consistent representation.
- weakly connected componentA weakly connected component is a maximal set of vertices in a directed graph that remains connected when edge directions are ignored.
- language modelA large language model is a neural language model trained on very large text corpora to understand and generate natural-language content.
- graph projectionA graph projection is an in-memory representation of selected graph data prepared for algorithmic analysis.
- domain knowledgeSpecialized knowledge and contextual understanding of a particular scientific or professional domain.
- entity resolutionEntity resolution is the process of determining which records or mentions refer to the same real-world entity.
- community detectionA community detection algorithm identifies groups of graph nodes that are more strongly or densely connected to one another than to nodes outside the groups.
- graph clustering methodA graph clustering method partitions graph nodes into groups according to structural or relational similarity.
← Chapter 3: Create your first knowledge graph from ontologies
Chapter 5: Extracting domain-specific knowledge from unstructured data →
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.