Pinned
Science doesn’t suffer from a lack of data.
It suffers from a lack of shared understanding.
The bottleneck isn’t computation.
It’s communication.
Nils Gehlenborg
274 posts
Nils Gehlenborg
@ngehlenborg
Faculty @harvarddbmi. Data Visualization. Biomedical Informatics. Genomics. Epigenomics. Cancer Biology. Single-Cell X. EHR UI. mHealth Data.
- The most important work feels boring to watch. It’s hard, slow, and rarely optimized for social media. But it’s the only work that lasts.
- Most infrastructure isn’t flashy. It’s silent, robust, and thankless until something breaks. That’s when you realize who built carefully, and who just built.
- We’re not short on data. We’re short on systems that make complexity understandable. The next frontier isn’t scale. It’s clarity.
- Biology is messy, chaotic, unpredictable. AI brings structure, prediction, and possibility. The intersection between the two is where the next generation of medicine will be born.
- We’re entering an era where AI isn’t just reacting to data. It’s starting to navigate systems. From biology to markets it’s not a chatbot revolution. It’s a systems revolution.
- The most dangerous assumption in science is that data alone is enough. What we see defines what we understand.
- The beauty of biology is in its messiness. The beauty of informatics is in helping us see patterns within that chaos —without flattening its complexity.
- 95% of meaningful progress happens behind closed doors, long before it becomes visible to the world. The ones who stay consistent without applause end up defining the future.
- This is a terrible situation! AI is the most transformative technology in human history, yet we are denying the best of the best the chance to work here. America-the land of the free & of immigrants-cannot maintain its competitive advantage if it keeps hurting itself with this
- It’s true that acceleration is undeniable. But I worry that while the tech jumps decades, our ability to interpret and interface with it still crawls. If we don't fix how humans interact with complexity, the future risks outrunning our understanding.We will see a century’s worth of advance in the next 10 years. AI, robotics, and domain advances in biotech, energy will compress decades into quarters. The next 10 years will feel less like “the 2030s” and more like flipping straight from 1925 to 2025 in one giant page‑turn.
- One particularly exciting direction for CellFlow is its potential integration with epigenomic data layers such as ATAC-seq, DNA methylation, or histone modification profiles. Since flow matching captures smooth, learnable trajectories of perturbation-induced cellular change, it1/ Excited to share CellFlow, a new approach for complex perturbation modeling in single-cell genomics based on flow matching. From cytokine screens to cell fate and organoid engineering, we show CellFlow’s broad power across many diverse tasks. 👉 Paper: biorxiv.org/content/10.110…
- The human brain didn’t evolve to understand 40,000 gene expression vectors. It evolved to recognize form, shape, and signal.
- We teach machines to read petabytes. But we still haven’t figured out how to help humans read a figure.
