Metastasis is the infiltrator. It is the means by which cancers start in one part of the body and suddenly end up elsewhere-- spreading destruction, and by simple statistics, enabling the genomic diversity that promotes drug resistance for the cancer. How does this happen? Well, it appears part of the problem starts, at least for head and neck cancer, with tumor cells initiating organizational changes in the lymph node wherein cancer has established itself, and initiated pro-cancer environments both locally and at a distance. Myeloid cells and cancer-associated fibroblasts are somehow instructed by distant cancer to "open the door" for seeding of the originating cancer-- which then further enables spread to nearby locales. In layman's terms... they prepare the ground elsewhere for their own defense. Cancer uses your own immune system against you. Congratulations to MH in my lab and to the many others from the collaborative University of Heidelberg & Mainz German Cancer Network who worked this out. The work offers new avenues of attack against such cancers by targeting cell-cell interactions beyond the current therapeutic framework. It also means we might be able to see, in non-metastatic lymph nodes, niche rearrangements that indicate cancer is nearby. It is now an open-source article online and in press at Cancer Cell. sciencedirect.com/science/ar…

There are hidden and embedded relationships in deep data that allow information transfer across knowledge domains (read that as RNA, protein, chromatic) even with "weakly linked" datasets. What does this mean? It means, to me, the end of worrying about running all 'omic interrogations on the exact same sample at the exact same time. It implies you can do a single-cell RNA seq on a comparable sample of the tissue and "transfer learn" data ONTO spatial imaging data with high accuracy. That means you can take deep data from high-fidelity scRNAseq and transfer the data onto, for instance, CODEX imaging (@AkoyaBio) with up to 90% accuracy. Thousands of RNAs "transferred" per cell versus maybe getting a few hundred RNAs if you were to try to measure the RNAs with a wet lab RNA imaging technique (often at low resolution anyway). Perfect? No... but better than not having the data at all. This implies a lot of single-cell genomic RNA and ATAC data sitting in databases can be harvested for probabilistic inference in imaging. The antibody-based imaging data becomes an "attractor" for the relevant RNA data. Work done with the amazing @ZongmingMa , @NancyZh60672287, @ShuxiaoC from UPenn, and Bokai Zhu from my lab.

Human languages have similar structures that describe the world. Researchers use the parallel languages of proteins, RNA, & epigenetics to describe cells/tissues. Such similarities can now fuse independent modalities w/ high fidelity. biorxiv.org/content/10.1101/…