Full article is here: nature.com/articles/s41551-0… and here: rdcu.be/dBZJK

We also generated 1M images for download: datadryad.org/stash/dataset/… Amazing work by Francisco Carrillo Pérez in the lab, and only possible thanks to the Polaris compute with Ravi Madduri at Argonne National Laboratory, U.S. Department of Energy (DOE) @madduri @argonne @ENERGY

We show that these synthetic data can be used in combination with real data, cell type distributions are representative of real tissues and synthetic data can be used for self supervised learning. You can try the model here: lnkd.in/egWGGDYJ

Finally, able to share our work in multi-modal synthetic data generation. We have developed a biomedical model inspired by DALL-E, we use RNA expression profiles to generate synthetic digital pathology images across several cancer tissues: rdcu.be/dBZJK

I was thrilled to get the invite from Russ Altman's podcast, The Future of Everything, to talk about our work in spatial omics applied to cancer patients and beyond: engineering.stanford.edu/mag… @StanfordEng

Our first foray in developing cross-modal biomedical data models to generate synthetic biomedical data. We developed RNA-GAN which uses RNA expression profiles as input to generate synthetic H&E images. shorturl.at/irwM9

Former postdoc in the Gevaert lab, Pritam Mukherjee developed using past medical history and lab results to predict future diagnoses. This work compares old vs. new AI methods. Read more here: rb.gy/xoo4f #ehr #machinelearning #artificialintelligence

here is a short video explaining the work: youtube.com/watch?v=7JxOaLAU…

Exciting work in glioblastoma research spearheaded by postdoc in the lab Yuan-Ning Zheng! Our team has developed a deep learning model to predict transcriptional subtypes of glioblastoma cells from spatial transcriptomics and histology images: nature.com/articles/s41467-0…

Thrilled to share the latest opportunistic screening work led by the brilliant @AyisPyrros for early type 2 diabetes diagnosis from routine chest x-rays. This work is yet another computer vision model that demonstrates the untapped potential for population level screening with imaging biomarkers using routine radiographic data! Check out our full study in Nature: rb.gy/4cya8

Great work together with Ghent University on developing whole slide imaging deep learning models to predict TP53 mutation for prostate cancer patients. we developed TiDo, a deep learning model that achieves state-of-the-art performance in predicting TP53…lnkd.in/gY4wSqA4

More details here: cell.com/cell-reports-method…

We just released our improved method for modeling joint DNA methylation and RNA sequencing data: EpiMix. Besides modeling protein coding genes, EpiMix models microRNAs, long non coding RNAs and enhancers and includes deconvolution methods for methylation & RNA sequencing data.

Published a few weeks ago and sharing here as well, a perspective with our thoughts on the importance of multi-modal data fusion in the context of cancer biomarker discovery. Full text access can be found here: rdcu.be/c9kEB. #datafusion #biomedical

We have developed a web-based app to provide personalized recommendations for COVID-19: covapp.stanford.edu/. You can read more about it here: nature.com/articles/s41591-0…

New work by superstar lab member Marie Humbert-Droz on extracting symptoms from clinical notes. Marie showed that deep learning approaches really shine with large cohorts: 1) performance becomes independent of prevalence & 2) validates in external cohorts. bit.ly/3qvBGsN

A long journey came to fruitful end with the work of Kevin Brennan now published in Human Molecular Genetics. Dr. Brennan studied the DNA methylation patterns in sotos syndrome, a common overgrowth with intellectual disability (OG…lnkd.in/gXwFACmS lnkd.in/gNfam_Bt

Stanford HAI is looking for Assitant Professor in their Junior Fellow program. This search is open to all fields in AI and Machine Learning across all Stanford schools & departments. Apply here bit.ly/3vozZP7

Great work by Hui Qu and Mu Zhou showing how deep learning models of whole slide images can be used to predict relevant molecular biomarkers of breast cancer patients. go.nature.com/3v4Q2l7