Abstract 4218: ProteoBridge: Bridging skipped sections via histology-based protein prediction

Key Points

• This research aims to develop ProteoBridge, a framework for predicting protein expression in histology slides to enhance 3D tumor reconstructions.
• Developed a deep learning model called ProteoBridge for protein pattern prediction.
• Trained the model using H&E images, multiplex protein images, and RNA transcriptomics from tissue sections.
• Focused on reducing mean absolute error to improve protein predictions across unmeasured sections.
• ProteoBridge accurately predicted marker intensities and spatial organization of proteins across tissue sections.
• It maintained consistency when filling in skipped sections, enhancing the continuity of 3D protein representation.
• The approach reduces costs and turnaround time for proteomic analysis while allowing for comprehensive profiling.