Neuro-Symbolic IA System Architecture

Neuro-Symbolic IA System Architecture

We present a novel approach based on the integration of Neuro-Symbolic AI systems. The symbolic system is implemented by deductive databases, enhancing the predictive capacity of subsymbolic systems implemented as Knowledge Graph Embedding models. As a proof of concept, we assess the performance of the proposed neuro-symbolic system on top of a KG of lung cancer treatments; the predictive task is to predict treatment effectiveness. The deductive database that specifies the rules relies on our previous work, “Capturing Knowledge about Drug-Drug Interactions to Enhance Treatment Effectiveness”, where we propose a deductive system over knowledge graphs to formalize the process of pharmacokinetic DDIs.

Running Neuro-Symbolic IA System

  • Descriptive_Analysis_Treatment_Response.ipynb: Shows a descriptive analysis of the treatment-response in the KG.
  • Analysis_DDIs_by_Treatment_Response.ipynb: Shows a detailed analysis of the distribution and density of DDIs by treatment-response in the benchmarks.
  • Statistics_TKG.ipynb: presents the metrics to measure size, diversity, and sparsity in Knowledge Graph.
  • Traverse_TKG.ipynb: You can traverse the benchmarks with your onw SPARQL queries.
  • deductive_system.py: presents the symbolic system implemented by a deductive database in datalog for the problem of treatment-response.
  • Analysis_BoxPlot_CosineSimilarity.ipynb: illustrates the analysis of cosine similarity in a box plot.
  • Evaluation_Integrated_SymbolicSubsymbolic_System.ipynb: illustrates the neuro-symbolic system evaluation metrics in the three benchmarks.
  • Score_Value_Predicted_Entities.ipynb: shows the behaviour of the scoring function for the entities predicted by each embedding model.
  • Link_Prediction.py: executes the link prediction task for the different negative sampling techniques and KGE models.