Integrating Knowledge Graphs

for Explainable Artificial Intelligence

in Biomedicine

Marta Contreiras Silva, Daniel Faria, Catia Pesquita

LASIGE, Faculdade de Ciências, Universidade de Lisboa

Abstract

Motivation

The current availability of public biomedical data and ontologies can be leveraged towards the development of explainable knowledge enabled systems in biomedical Artificial Intelligence, with the objective of developing personalized medicine solutions to aid clinicians and researchers.

We propose an approach to build a Knowledge Graph (KG) for personalized medicine to serve as a rich input for the AI system and incorporate its outcomes to support explanations, by connecting input and output.

Personalized Medicine

is a data-oriented medical model that relies on the specific characteristics of the patient and thus encompasses a very diverse and complex domain

Explainable Artificial Intelligence (XAI)

is able to handle the vast biomedical data pool available and has the goal of providing human-understandable explanations in order to increase trust in AI model predictions

Knowledge Graphs

are able to combine biomedical ontologies and data to provide a rich and complex domain context, functioning as input for biomedical XAI applications capable of producing explanations

Our approach

We are building a KG to support XAI in personalised cancer therapy using semi-automated methods for linking publicly available biomedical ontologies and datasets. Ontologies can be integrated iteratively through ontology matching to create the semantic backbone of the KG, which can then be populated with public datasets through semantic data annotation. The resulting KG will be used as input by AI approaches, and moreover can integrate AI outcomes to support explanations.

Resource curation

Selection and curation of relevant public biomedical datasets and ontologies

 

Challenges:

  • Adequate coverage of the domain
  • Semantic richness for explanations

Ontology Matching

Ontologies will be matched iteratively and alignments will be partially validated by experts

 

Challenges:

  • Scalability
  • Complex matching

Semantic Data Annotation

Development of parsers and annotation algorithms

 

Challenges:

  • Diverse types of datasets

AI Integration

Ensure that the KG serves as both input to AI methods and also encodes the AI outcomes

 

Challenges:

  • Diverse types of AI outcomes

Conclusions

  •    A biomedical KG obtained from the intersection of biomedical ontologies and data can be leveraged as input to AI methods and can encode its outcomes
  •    Such a KG can support a shared semantic space for data, scientific context, and predictions capable of supporting KG-based explanation methods

Authors

Marta Contreiras Silva

LASIGE, Faculdade de Ciências

Daniel Faria

LASIGE, Faculdade de Ciências

Catia Pesquita

LASIGE, Faculdade de Ciências

Funding

This work was supported by FCT through the LASIGE Research Unit (UIDB/00408/2020 and UIDP/00408/2020). It was also partially supported by the KATY project which has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 101017453.