Finally, this work provides inspiration and use cases for our machine learning tools in @galaxyproject.bsky.social
, which is part of #NCIITCR These tools simplify machine learning and promote best practices when using machine learning for cancer and other biomedical research. 8/ End 🧵

Finally, we used imputed protein abundances levels to accurately predict treatment timepoints for single cells. This demonstrates the biological applicability of our methods by providing insight into the cellular populations present before and after therapy. 7/

Using spatial information improves predictive performance of the machine learning models, especially for proteins that are highly variable. This is very important as highly variable proteins are often quite prevalent throughout a tissue sample 6/

Deep learning approaches enable predictions for multiple proteins at once with accuracy that is similar to machine learning models dedicated to predicting a single protein 5/

Error rates for predicted proteins are low overall, and multiple machine learning methods can be used to make accurate predictions. 4/

In this work, we were able to accurately predict protein abundance levels in single-cell multiplex tissue imaging (t-cycIF) datasets from breast cancer. These predictions enable "virtual imaging" and obtaining abundance levels when assays fail or do not image for a protein 3/

This work was led by the very talented @ohsunews.bsky.social Ph.D. student Raphael Kirchgaessner with contributions from others not on social media. This work is also part of the NCI #HTAN 2/

#NCIITCR tools used at the workshop include
@galaxyproject.bsky.social, @bioconductor.bsky.social, spatialGE, and spatialTIME Thank you
@itcrtraining.bsky.social