Deep learning for Multimodal Data Integration

Book chapter in Circulating Tumor Cells, 2nd Edition

---

Abstract

The complex multimodality in clinical and liquid biopsy data generated by multiparametric diagnostic and tumor profiling technologies presents an exciting opportunity for developing innovative predictive mathematical models by harnessing large datasets across studies and institutional data repositories. Further, comprehensive liquid biopsy analysis with single-cell profiling provides multiscale data on the morphology, genomics, and proteomics of circulating tumor cells (CTCs) and tumor microenvironment cells with deeper resolution on spatiotemporal tumor biology. However, clinical datasets are frequently sparse, inconsistent, and incomplete, due to the lack of standardization and interoperability across cancer centers and healthcare systems. In liquid biopsies where specimens are oftentimes samples of convenience, datasets are often of limited scale and lacking paired clinical data. Methods for integrating multimodal oncology data are paramount for building robust models for outcomes prediction. This chapter presents methodological challenges and advances in handling missingness and sparsity, integrating multidimensional clinical and multi-omic liquid biopsy data for improving the accuracy and robustness of machine and deep learning survival prediction models. We highlight the critical role that the convergence of artificial intelligence and liquid biopsies holds for the future of therapy response and survival prediction in precision oncology.

Read the book chapter in Circulating Tumor Cells, 2nd Edition

Reference:

Ndacayisaba, L.J., Mason, J., Kuhn, P. (2023). Mathematical Oncology to Integrate Multimodal Clinical and Liquid Biopsy Data for the Prediction of Survival. In: Cote, R.J., Lianidou, E. (eds) Circulating Tumor Cells. Current Cancer Research. Springer, Cham. doi.org/10.1007/978-3-031-22903-9_7