Multimodal Approach for Diagnosis and Prognosis of Metabolic and Alcohol-Associated Steatotic Liver Diseases

PROJECT SUMMARY Alcohol-associated liver disease (ALD) and metabolic dysfunction-associated steatotic liver disease (MASLD) are leading causes of liver transplantation, morbidity, and mortality. With the obesity epidemic and increasing alcohol consumption, a mixed-etiology disease, metabolic syndrome-associated alcoholic liver disease (MetALD), has emerged. Accurate diagnosis and risk prediction of ALD, MASLD, and MetALD are currently hindered by the lack of distinct histological features. To address this critical knowledge gap and its impact on diagnostic accuracy and the development of precision medicine, this proposal introduces an innovative approach that harnesses the power of multi-omics data integration using machine learning (ML) techniques. Our central hypothesis is that each disease has distinctive features identifiable by ML and spatial transcriptomics, enabling the development of new diagnostic and prognostic tools. The rationale for this is that current histological methods lack specific features, whereas ML has shown potential for enhanced diagnostic accuracy. In our preliminary data: 1) we developed several ML tools, including one that distinguishes between ALD and MASLD with high accuracy, another that assigns zones in liver biopsy whole slide images (WSI), and identified a morphometric signature for severity in ALD; 2) we identified a unique zonal gene signature profile for MetALD from a limited spatial transcriptomic dataset. Building on these preliminary successes, our proposal outlines two specific aims designed to further dissect the complexities of steatotic liver disease: 1. We will define the etiology and predict the severity of steatohepatitis using novel ML tools as follows: a. Classify the etiology as ALD, MetALD, or MASLD on WSI based on ML. b. Predict disease severity using a graph neural network model for comprehensive morphometric analysis. 2. We will identify distinct profiles that confer prognostic value for MetALD as follows: a. Define zonal and unique regions of interest pathways of liver injury using spatial transcriptomics b. Develop a multimodal prognostic algorithm for MetALD based on the histological features of the graph neural network, clinical and serological findings, and spatial transcriptomics. This proposal is technically and conceptually innovative, as it seeks to integrate histological features with ML for diagnostic precision and disease severity prediction, advancing our understanding of the underlying pathways of injury in MetALD. These outcomes will guide the development of noninvasive diagnostic and prognostic markers, aligning with NIAAA’s mission to improve alcohol-related liver disease outcomes. This proposal ensures that I achieve my goal of becoming an independent physician-scientist specializing in ML, spatial transcriptomics, and liver diseases, through the advanced training and dedicated research time provided by the K08 and the well-rounded mentorship and tremendous resources provided by the Mayo Clinic.