This repository consists of source files for the implementation of a hybrid AI (HyAI) framework. We have not publicly available the HyAI Knowledge Graph (KG) due to sensitive information about chronic hepatitis B virus (HBV) infected patients. HyAI is conceptualized following the design principles described by Bekkhum, V, et al. Design Patterns; following basic vocabulary for representing the components actor, input and output, process, and models (as depicted in the below Figure). HyAI framework consists of four design patterns: (i) Ontology and KG, (ii) KG Embedding, (iii) Pattern Detection, and (iv) Pattern Analysis and Explanation.

We used HyAI in the use case of uncovering parameters of clinical, demographic, and immune phenotyping data that characterize chronic HBV patients with functional cure. HyAI has been implemented using state-of-the-art tools and techniques (as depicted in the below Figure). The heterogeneous datasets consists of 87 chronic HBV patients, including age, sex, 18 clinical observational parameters, 45 immune phenotyping parameters, and HBV treatment. The Ontology and KG system received a data integration system (DIS), that is composed of a unified schema (classes and properties), data sources, and RML mapping assertions. The KG embedding models (TransE, TransH, RESCAL, ERMLP) have been used to transform holistic profiles of 87 chronic HBV patients into low-dimensional vector representations. The Pattern Detection system used community detection algorithms (KMean, SemEP, METIS) to identify groups of HBV patients who shares similar features.

HyAI captures knowledge encoded in chronic HBV-infected patients during experimental setup. The quality metrics used in the experimental setup for community detection are (i) Inverse Conductance (InvC), (ii) Inverse Total Cut (InvTC), and (iii) Coverage (Co), using algorithms (SemEP, METIS, and KMeans). The values higher for InvC, InvTC, and Co are considered better. Figure b, c, d, and e assess HyAI framework, while Figure a shows baseline measurement. These experiments observed that the communities’ quality generated by HyAI performs better than the baseline.

• GNU Compiler Collection (GCC) or Clang
• GNU make
• pykeen
• pandas
• numpy
• scipy
• seaborn
• sklearn

python KGEmbedding/ComputeKGE.py models/ ./HyAI_KG.csv

python PatternDetection/ComputeCommunities.py

python PatternDetection/evaluation_metric.py

• StatisticsHyAI_KG.ipynb: presents the metrics to measure size in HyAI KG.
• BaselineRelationalForm.ipynb: we establish a baseline using the HBV data in its relational form.
• JaccardIndex_CategoryBasedScore.ipynb: shows the quality of the communities based on the gold standard. Metrics $\mathcal{CS}$ and $\mathcal{J}$ assess the baseline and HyAI.
• ViolinPlot.ipynb: Analyzing Patterns of HBV Patients. HyAI Partitioning.

If you find HyAI helpful in your work please cite the paper:

Shahi Dost, Ariam Rivas, Hanan Begali, Annett Ziegler, Elmira Aliabadi, Markus Cornberg, Anke RM Kraft  and Maria-Esther Vidal. 2023.
Unraveling the Hepatitis B Cure: A Hybrid AI Approach for Capturing Knowledge about the Immune System’s Impact (ACM ISBN 979-8-4007-0141-2/23/12)[https://doi.org/10.1145/3587259.3627558]

HyAI codes and source files are licensed under MIT License Copyright (c) 2023.