Knowledge-infused Global-Local Data Fusion for Spatial Predictive Modeling in Precision Medicine.

Area of Science:

• Computational biology
• Medical imaging
• Machine learning

Background:

• Precision medicine aims to tailor cancer treatments using molecular markers.
• Tumor molecular markers exhibit significant spatial heterogeneity, complicating treatment.
• Accurate spatial prediction of these markers is crucial for personalized therapy.

Purpose of the Study:

• To develop a novel machine learning framework for spatial prediction of molecular markers in tumors.
• To fuse sparse biopsy data, global imaging data, and mechanistic models for enhanced prediction.
• To improve the precision of cancer treatment by accounting for spatial heterogeneity.

Main Methods:

• Proposed a knowledge-infused global-local data fusion (KGL) model.
• Developed and theoretically studied a novel mathematical formulation for data fusion.
• Applied the KGL model to predict Tumor Cell Density (TCD) in glioblastoma using biopsy, MRI, and a PDE simulator (Proliferation-Invasion model).

Main Results:

• The KGL model demonstrated superior prediction accuracy for Tumor Cell Density compared to existing methods.
• KGL achieved the minimum prediction uncertainty in spatial distribution predictions.
• Real-data application showed the model's effectiveness in handling complex tumor data.

Conclusions:

• The KGL model offers a robust approach to fuse multi-modal data for accurate spatial prediction.
• This framework has significant implications for developing individualized and spatially-optimized cancer treatments.
• Accurate spatial prediction of tumor markers can enhance therapeutic strategies in precision oncology.