So fundamental to making these discoveries was the use of the TRACERx renal data set. So this is a very special data set in which we take multiple samples from ex vivo nephrectomy specimens in clear cell renal cell carcinoma. And we take multiple samples and each one is divided into two to provide a very close link between the histology and the genetic data. Then we can go on to do unsupervised clustering and Bayesian logistic regression models to find links between the histological features and the genetic sequencing data...
So fundamental to making these discoveries was the use of the TRACERx renal data set. So this is a very special data set in which we take multiple samples from ex vivo nephrectomy specimens in clear cell renal cell carcinoma. And we take multiple samples and each one is divided into two to provide a very close link between the histology and the genetic data. Then we can go on to do unsupervised clustering and Bayesian logistic regression models to find links between the histological features and the genetic sequencing data. So in terms of the histological data, we had a three-pronged approach. The first was a pathologist, that would be my part in the study, annotating on different histological architectural and pathological features. And we also used vascular topology segmentation pipelines and unsupervised or convolutional neural network-based models to predict the genetic alterations directly from the H&E slide. What we found was that genetic alterations do leave morphological imprints in the tissue morphology, and critically, these change with accumulating driver events. So, for instance, there’s one mutation called PBRM1, which is associated with a very specific set of histological features when it occurs by itself, with additional genetic driver events accumulating, the appearances of that change. And so histology actually tracks alongside clonal evolution.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
