Photoaging Skin
Edwardia Fosah
Lay Summary:
In my project, I set out to create high quality, digital annotations for skin tissue data. These annotations could be used in Levy Lab to train segmentation and classification models and validate the ability for inferred spatial gene expression data to be used to predict the cell type abundances annotated in the project.
Abstract:
According to the American Academy of Dermatology Association, skin cancer is the most common form of cancer in the United States, and with this high incidence rate, accurate measurement of its major risk factors is crucial in improving patient outcomes through early cancer detection. Photoaging, histological and molecular skin damage resulting from acute UV radiation, is a major risk factor for skin cancer but lacks reliable and quantifiable forms of measurement. Virtual RNA Inference (VRI), a computational method that uses hematoxylin and eosin (H&E) stained tissue images to predict spatial gene expression without the use of direct measurement as with spatial transcriptomics (ST), could be used to identify photoaging features in cells with comparable performance to ST technologies. Additionally, the inferred ST data from VRI could be applied to downstream tasks such as cell type abundances prediction and age prediction. In this, we expect inferred ST data to be able to predict age and cell type abundance at a comparable performance to measured ST data by Visium on skin tissue histologies. During this research, 9 different UV-damaged skin tissues collected at Dartmouth Health Dermatology in the Mohs Clinic were annotated with 1228 individual segmentations of sebaceous glands, hair follicles, vessels, eccrine glands, nerves, fat, epidermis, and smooth muscles. QuPath was used for all annotations, and all annotations were exported as GeoJSONs, FeatureCollections, and with measurements. These annotations can be used to train future segmentation and patch-level classification models and validate the application of inferred ST data on downstream tasks. More expansive research that includes more areas of anatomy would be required to create a framework that can generate accurate spatial gene expression data across tissue domains.
Q&A:
Bios: Edwardia Fosah
Program Track: Advanced Research
GitHub Username:
edwardiafos -Edwardia Fosah
What was your favorite seminar? Why?
My favorite seminar has to be the seminar about writing and preparing manuscripts because it found it to be the most useful for my internship and extremely insightful to the true purpose of scientific research: communication. -Edwardia Fosah
If you were to summarize your summer internship experience in one sentence, what would it be?
I learned immensely about skin histology, QuPath annotating, and spatial transcriptomics/adjacent technologies. -Edwardia Fosah