Dr. Jean Fan

Excited to share a substantial update to our preprint on STARIT (Spatial Transcriptomics As Rasterized Image Tensors), a framework for enabling computer vision AI analysis of spatial transcriptomics data for characterizing subcellular heterogeneity. Preprint: v2 www.biorxiv.org/content/10.6...

Using real imaging-based spatial transcriptomics (MERFISH) data of bacteria, STARIT identifies cells with similar overall exp magnitude but distinct subcellular organization: showing more punctuated vs diffuse transcript localization for flagellar subunits, with potential functional consequences.

Excited to share a substantial update to our preprint on STARIT (Spatial Transcriptomics As Rasterized Image Tensors), a framework for enabling computer vision AI analysis of spatial transcriptomics data for characterizing subcellular heterogeneity. Preprint: v2 www.biorxiv.org/content/10.6...

Imaging-based spatial transcriptomics is often represented as cell-level gene counts, overlooking subcellular spatial variation. STARIT creates an image-based representation, allowing us to use computer vision AI models to identify cell states with distinct subcellular localization patterns.

We've updated our preprint characterizing the spatiotemporal impact of cold ischemia in mouse kidney to better understand the molecular underpinnings of cold-storage injury in kidney transplants. Check it out: www.biorxiv.org/content/10.1... 🧵👇

Our spatiotemporal transcriptomic analysis thus identified coordinated molecular changes within metabolic pathways deep within the cold ischemic kidney, highlighting potential opportunities for new insights beyond those available from superficial biopsy-focused tissue examinations.

We've updated our preprint characterizing the spatiotemporal impact of cold ischemia in mouse kidney to better understand the molecular underpinnings of cold-storage injury in kidney transplants. Check it out: www.biorxiv.org/content/10.1... 🧵👇

Increased durations of cold ischemia is implicated in poor transplant outcomes, but the molecular mechanisms remain unclear. Using spatial transcriptomics, we profiling cold ischemia injury using a mouse model to identify temporally-resolved gene expression changes in a compartment-specific manner.

As an alternative to the h-index, I made the Mentorship Index (M-index) to proxy a scientist's contribution to mentoring junior scientists. Ex. M10-index = # last-author publications where the first author had < 10 pubs. Calculate yours: jef.works/Mentorship-I... Blog: jef.works/blog/2026/03... 🧵👇

The M-index isn’t meant to replace holistic evaluation of mentoring abilities. But such metrics can help surface something I care about in ways that existing metrics like the h-index doesn’t capture. If we don’t create metrics that reflect our values, we cede that ground to metrics that may not.

As an alternative to the h-index, I made the Mentorship Index (M-index) to proxy a scientist's contribution to mentoring junior scientists. Ex. M10-index = # last-author publications where the first author had < 10 pubs. Calculate yours: jef.works/Mentorship-I... Blog: jef.works/blog/2026/03... 🧵👇

My approach: I use openalex.org to count the # of publications for which the scientist served as last author (senior/mentoring role) where the first author (the mentee who led the work) was relatively new to science (proxied by the # of publications associated with their name at that time).

As an alternative to the h-index, I made the Mentorship Index (M-index) to proxy a scientist's contribution to mentoring junior scientists. Ex. M10-index = # last-author publications where the first author had < 10 pubs. Calculate yours: jef.works/Mentorship-I... Blog: jef.works/blog/2026/03... 🧵👇

We're recruiting senior faculty! So at faculty meeting, a colleague brought up a candidate who has an h-index of 100. Impressive! But then another colleague pointed out: their last senior-author paper was 10 years ago. This got me thinking about what metrics actually capture and what they miss.

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

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

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

In this blog post, I vibe code to apply STcompare to kidneys assayed by two different spatial transcriptomics technologies (Visium vs. Xenium) to identify spatially consistent and differentially patterned genes. Follow along + try it out for yourself: jef.works/blog/2026/02...

We recently updated our paper demonstrating evidence of off-target probe binding affecting the 10x Genomics Xenium spatial transcriptomics platform with key clarifications, new quantifications, and approaches for evaluating custom gene panels: biorxiv.org/content/10.1... 🧵👇1/n

We thank the editors and reviewers at eLife for their feedback. You can check out their public peer review comments here: elifesciences.org/reviewed-pre...

We recently updated our paper demonstrating evidence of off-target probe binding affecting the 10x Genomics Xenium spatial transcriptomics platform with key clarifications, new quantifications, and approaches for evaluating custom gene panels: biorxiv.org/content/10.1... 🧵👇1/n

Finally, we emphasize that such off-target binding may similarly affect other probe-based gene detection approaches from other commercial vendors given sensitivity vs. specificity tradeoffs. We therefore emphasize the importance of transparency through sharing probe sequences.

We recently updated our paper demonstrating evidence of off-target probe binding affecting the 10x Genomics Xenium spatial transcriptomics platform with key clarifications, new quantifications, and approaches for evaluating custom gene panels: biorxiv.org/content/10.1... 🧵👇1/n

We now clarify the differences between these probe sequences in a new supplementary note. Importantly, this impacts our interpretation of observed results to provide stronger evidence of putative imperfect sequence homology based off-target probe binding.

We recently updated our paper demonstrating evidence of off-target probe binding affecting the 10x Genomics Xenium spatial transcriptomics platform with key clarifications, new quantifications, and approaches for evaluating custom gene panels: biorxiv.org/content/10.1... 🧵👇1/n