Deep Phenotyping FSGS Using Artificial Intelligence and Special Transcriptome
Abstract
Focal segmental glomerulosclerosis (FSGS) is a heterogeneous clinicopathological condition that can be classified into four categories according to etiology: (1) primary FSGS attributed to circulating permeability factor, (2) genetic FSGS caused by gene mutations, (3) secondary FSGS associated with, e.g., viral infections, medication or maladaptive changes, and (4) FSGS occurring without an identified cause. Some FSGS patients respond promptly to steroids; however, approximately 50% of individuals progress to end-stage kidney disease within 5-10 years despite therapy. FSGS is diagnosed by the morphologic changes observed in the renal biopsy sample. Currently, the Columbia classification used in renal pathology does not match well with the etiology and is not sufficiently robust to guide treatment and predict the clinical prognosis. The development of a precise and prognosis-relevant FSGS classification is limited by two factors: empirical morphometric features without quantitative molecular evidence, and biased and less reproducible semi-quantitative assessment of these features. Artificial intelligence (AI) has been applied to medical images, including histopathology, from which AI could extract more prognostic information than an experienced pathologist. Our approach is to develop a quantitative deep phenotyping algorithm for etiology-related signatures identification and quantification in FSGS. Spatial transcriptomics allows us to measure all the gene activity and map their locations in the histological sections. By using this spatial transcriptome-supervised machine learning and quantitative AI algorithms, we will detect and quantify those etiology-related signatures in the FSGS biopsy samples. We will train the AI in mouse FSGS models, and transfer and apply this AI algorithm to human samples. This strategy will significantly reduce the requirement of a human training set. The quantitation of these morphology signatures will be carried on in a fraction of serial sections of renal biopsy, which show a more accurate prediction of FSGS lesions than 2D-based quantification. This approach will help to generate a reference FSGS atlas by combining regular staining, spatial transcriptome, and computer vision in the feature. A new classification based on these etiology and transcriptomic related spatial signatures will not only improve the precision and reproducibility of diagnosis but also facilitate the identification of specific drug targets and ultimately enable individualized care in FSGS.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2024
- Source ID
- HT94252310003
Entities
People
- Haichun Yang
Organizations
- United States Army
- Vanderbilt University