...
Please join us for our next CPTAC Imaging SIG webinar at Monday, Sep 9, 2019 12:00 pm Eastern. In this session we'll have two speakers who will provide summaries of summarize their work using CPTAC data in imaging-omic correlation studieslinking the imaging, genomics and proteomics data from CPTAC patients.
1) Dr. Olivier Gevaert is an assistant professor at Stanford University focusing focused on developing machine-learning methods for biomedical decision support from multi-scale data. His lab focuses on multi-scale biomedical data fusion primarily in oncology and neuroscience. The lab develops machine learning methods including Bayesian, kernel methods, regularized regression and deep learning to integrate, clinical, molecular and biomedical image data. His presentation will show an example of how to process proteomic data from CPTAC Phase 2 projects (breast, ovarian and colorectal) with emphasis on how to use, preprocess and subsequently model proteomic data using bioinformatics algorithms. He will show an example of linking protein data to DNA methylation and mRNA gene expression data, and how proteomic data can be integrated with medical image data.
2) Runyu Hong is a Graduate Assistant/SCBM PhD Student at the Fenyö Lab in the computational biomedicine PhD student working in Dr. David Fenyö’s lab of NYU School of Medicine’s Institute for Systems Genetics at the NYU Langone Health/NYU School of Medicine. He will speak about their project which trained a deep learning model that can to distinguish STK11 mutated and wild type pathology slides from CPTAC-LUAD (lung adenocarcinoma) patients. The STK11 gene provides instructions for making a tumor suppressor, serine/threonine kinases 11. Multi-omics analyses of CPTAC non-small-cell lung cancer (LUAD. He will also discuss how they were able to visualize the morphological features correlated with STK11 mutations based on this model) datasets have found that STK11-mutated patients showed less immune response than other patients. To visualize and validate this pattern in histopathology images, they trained an InceptionV3-architectured convolutional neural network model that can achieve high performance (AUC=0.94444) in predicting STK11 mutations based on histopathology images. By extracting the model’s last layer activation maps of a random-sampled tiled pieces of images from the test set, they clustered these tiles with dimensional reduction method for features’ visualization. An experienced pathologist examined the tiles in the positively and negatively predicted clusters and was able to conclude that tiles in the STK11-postive clusters generally show plenty of cancer cells, but very few immune cells compared to the ones in STK11-negative clusters. This experiment supports the finding in multi-omics analyses and suggests that their model used immune cells as an important feature to distinguish STK11 mutated images.
Connection details:
URL: https://cbiit.webex.com/cbiit/j.php?MTID=m46af7437b801588996cf7f73600c8d9e
Phone: 650-479-3207
Access code: 736 153 817
...