Integration of statistical methodology with machine learning biomedical technologies and mixed effects machine learning for spatially localized omics information.
Our team has identified a number of statistical challenges which motivate further exploration into the development and application of hierarchical Bayesian methods for the development and evaluation of artificial intelligence technologies in digital pathology. To this end, we have developed software and methods which can extract findings from machine learning models, in the scenario where statistical dependencies may exist between observations, to be used in statistical models for clinical findings. The software has been used by members of the biomedical community. We have applied such methods to identify statistically significant interactions from spatially localized omics information for colorectal cancer metastasis prediction. We are planning follow-up works which will communicate clinical findings and motivate classifier deployment in this domain. Furthermore, we have developed a new class of Bayesian statistical methodology, Bridge Category Models, implemented in Stan, to account for phenomena where pathologists may report multiple stages at a time, which may reflect true clinical practice but underreported in clinical literature due to the difficulties associated with handling such data.
- Levy, J. J., Bobak, C. A., Nasir-Moin, M., et al. Mixed Effects Machine Learning Models for Colon Cancer Metastasis Prediction using Spatially Localized Immuno-Oncology Markers. Pacific Symposium on Biocomputing, 2022. PMID: 34890147
- Levy, J., Bobak, C., Azizgolshani, N., et al. Bridge Category Models: Development of Bayesian Modelling Procedures to Account for Bridge Ordinal Ratings for Disease Staging. 2021.08.17.456726 https://www.biorxiv.org/content/10.1101/2021.08.17.456726v2 (2021).
- Levy, J., Bobak, C., Azizgolshani, N., et al. Estimating the Inter- and Intra-Rater Reliability for NASH Fibrosis Staging in the Presence of Bridge Ordinal Ratings with Hierarchical Bridge Category Models. 2021.10.27.466144 https://www.biorxiv.org/content/10.1101/2021.10.27.466144v1 (2021).
- Levy, J., Bobak, C., Azizgolshani, N., et al. Improving the Virtual Trichrome Assessment through Bridge Category Models. 2021.10.30.466613 https://www.biorxiv.org/content/10.1101/2021.10.30.466613v1 (2021) doi:10.1101/2021.10.30.466613.
- Levy, J. J. & O’Malley, A. J. Don’t dismiss logistic regression: the case for sensible extraction of interactions in the era of machine learning. BMC Med Res Methodol 20, 171 (2020). PMCID: PMC7325087
- Levy, J., Bobak, C., Christensen, B., et al. Estimating latent positions in social and biological networks using Graph Neural Networks in R with GCN4R. 2020.11.02.364935 https://www.biorxiv.org/content/10.1101/2020.11.02.364935v2 (2021) doi:10.1101/2020.11.02.364935.