Post

Dr. Jean Fan
@jef.works

We previously showed how changing training data alone can improve deep learning prediction of spatial transcriptomics gene expression from histology images by +38% (without any changes to model architecture). We've now updated our preprint w/ expanded results: www.biorxiv.org/content/10.1... 🧵1/n

2 reposts3 likes

Dr. Jean Fan
@jef.works

We previously showed how improving data quality (molecular detection sensitivity, imaging resolution, etc) provides an orthogonal strategy to tuning model architecture in spatial transcriptomics-based predictive modeling. But data imputation intended to improve data quality led to overfitting. 2/n

Dr. Jean Fan
@jef.works

We've now updated our preprint to expand on these results to further demonstrate that these data quality-driven effects are reproducible across: ✅ More spatial transcriptomics datasets ✅ More spatial transcriptomics technologies ✅ Different feature extractors ✅ Alternative model architectures 3/n

Dr. Jean Fan
@jef.works

Check out our preprint with more details, quantifiable metrics, insufficiencies of data imputation, and other insights: www.biorxiv.org/content/10.1... Work led by Caleb Hallinan 🥳 4/4 Previous thread: bsky.app/profile/jef....

biorxiv.orgbiorxiv.org
Dr. Jean Fan@jef.works

High costs motivate efforts to predict spatial transcriptomics from H&E images w/ deep learning. In our recent preprint, we show that noise, sparsity & resolution in ST data impact performance, highlighting the importance of training data quality: www.biorxiv.org/content/10.1... 🧵1/n