Loading video...

Video Failed to Load

Go Home

Single-cell technologies now let us profile entire transcriptomes in individual cells. But how do we make sense of this complexity in a biologically meaningful way? Many methods summarise cells into a single embedding, but this often comes at the cost of interpretability, especially when multiple gene programs are active...

mo_lotfollahi's profile picture

Mo Lotfollahi

13,007 subscribers

22,578 views • 7 months ago •via X (Twitter)

Health & WellnessScience & Technology
LIVE

Anya Rossi• Live Now

Private livecam show

7 Comments

Kaven's profile picture
Kaven7 months ago

AI link for summary and cited qa

Mo Lotfollahi's profile picture
Mo Lotfollahi7 months ago

Ahh this is beautiful 😊

Kaven's profile picture
Kaven7 months ago

I love them, you can generate these editable pptx on nouswise for free.

Dr. Zeynep Hülya Gümüş 🕊️'s profile picture
Dr. Zeynep Hülya Gümüş 🕊️7 months ago

Congrats, Mo! 🔥

Jamie Timmons's profile picture
Jamie Timmons7 months ago

That first line isn't really true. They provide a partial, part biased, part stochastic, SAMPLE of the transcriptome - and the impact of these limitations is as of yet isn't fully understood.

Aiden M Sababi's profile picture
Aiden M Sababi7 months ago

Great work, congrats.

Jaime's profile picture
Jaime7 months ago

@HaoYin20 Clean approach — representing cells through multiple gene programs instead of a single embedding makes a lot of sense. Could the key question be not just which programs are active, but whether they stay dynamically accessible under perturbation?

Related Videos

Excited to share our new work on building a multimodal atlas of human skin in health and inflammatory disease — a project I’m especially proud of, bringing together AI, high-throughput genomics, and clinical science to accelerate discovery. Over the past decade, single-cell genomics has transformed how we map cells in human tissues. But a major challenge remains: can we systematically decode how cells organize into functional niches in situ — including those invisible to standard histopathology? To address this, we integrated large-scale scRNA-seq, spatial transcriptomics, histopathology, and AI-driven modeling frameworks to build an in situ atlas of human skin across health and disease. Led by Lloyd Steele, an MD/PhD student working between Haniffa Lab and my lab at Wellcome Sanger Institute and Cambridge University . Another amazing collaboration with Muzz Haniffa, the mastermind behind the work as part of Human Cell Atlas. A key part of this study is that we didn’t build everything from scratch — we leveraged and combined AI methods that actually work! and showed how they can be used together to extract biological insight at scale. We used: • scArches to build and map into a reference scRNA-seq atlas of human skin: • NicheCompass to identify and characterize spatial niches: • MINT-Flow to extract microenvironment-induced cell states and gene programs: Together, these enabled an end-to-end workflow from atlas construction to spatial mapping, niche discovery, and cell state decoding. At scale, we integrated ~5 million cells and 100+ spatial sections, enabling a systematic view of tissue organization. Using this framework, we identified 26 niches in skin, including known histopathologic structures as well as hidden disease-associated niches not visible on H&E. Among the most striking findings were a resident memory T cell-rich sebaceous gland niche and a plasma cell-rich sweat gland niche, suggesting that appendageal structures act as active immunological microenvironments and may contribute to inflammatory memory and disease persistence. Importantly, this atlas is not just descriptive — it is usable. It can support mapping of new datasets, resolve finer cell types and niches, extract microenvironment-driven programs, and enable predictive analyses at scale. More broadly, this work shows what becomes possible when AI, spatial genomics, and atlas-scale data are integrated end-to-end: not just mapping tissues, but systematically decoding them. This was a massive collaboration, and I’m very grateful to the amazing scientists April Foster, Kenny Roberts, and Chloe Admane. Lloyd is an amazing scientist, and I’m especially excited for the community to see more of his work soon — stay tuned. The data and pre-trained models will be released soon. Preprint:
1:09

Excited to share our new work on building a multimodal atlas of human skin in health and inflammatory disease — a project I’m especially proud of, bringing together AI, high-throughput genomics, and clinical science to accelerate discovery. Over the past decade, single-cell genomics has transformed how we map cells in human tissues. But a major challenge remains: can we systematically decode how cells organize into functional niches in situ — including those invisible to standard histopathology? To address this, we integrated large-scale scRNA-seq, spatial transcriptomics, histopathology, and AI-driven modeling frameworks to build an in situ atlas of human skin across health and disease. Led by Lloyd Steele, an MD/PhD student working between Haniffa Lab and my lab at Wellcome Sanger Institute and Cambridge University . Another amazing collaboration with Muzz Haniffa, the mastermind behind the work as part of Human Cell Atlas. A key part of this study is that we didn’t build everything from scratch — we leveraged and combined AI methods that actually work! and showed how they can be used together to extract biological insight at scale. We used: • scArches to build and map into a reference scRNA-seq atlas of human skin: • NicheCompass to identify and characterize spatial niches: • MINT-Flow to extract microenvironment-induced cell states and gene programs: Together, these enabled an end-to-end workflow from atlas construction to spatial mapping, niche discovery, and cell state decoding. At scale, we integrated ~5 million cells and 100+ spatial sections, enabling a systematic view of tissue organization. Using this framework, we identified 26 niches in skin, including known histopathologic structures as well as hidden disease-associated niches not visible on H&E. Among the most striking findings were a resident memory T cell-rich sebaceous gland niche and a plasma cell-rich sweat gland niche, suggesting that appendageal structures act as active immunological microenvironments and may contribute to inflammatory memory and disease persistence. Importantly, this atlas is not just descriptive — it is usable. It can support mapping of new datasets, resolve finer cell types and niches, extract microenvironment-driven programs, and enable predictive analyses at scale. More broadly, this work shows what becomes possible when AI, spatial genomics, and atlas-scale data are integrated end-to-end: not just mapping tissues, but systematically decoding them. This was a massive collaboration, and I’m very grateful to the amazing scientists April Foster, Kenny Roberts, and Chloe Admane. Lloyd is an amazing scientist, and I’m especially excited for the community to see more of his work soon — stay tuned. The data and pre-trained models will be released soon. Preprint:

Mo Lotfollahi

11,865 views • 7 months ago

Excited to share our new work. Over the past decade, single-cell genomics has transformed our ability to map cellular systems. But a major question remains: Can we predict how perturbations reshape cellular trajectories over time? In 2018, we first showed that it is possible to predict cellular responses to perturbations — ranging from disease signals to chemical treatments — even in unseen contexts. In 2022, we introduced CPA (MSB 2022; NeurIPS 2022), extending this idea to predict responses to unseen chemical and genetic perturbations, including their combinations. Since then, the field of perturbation modeling has grown enormously. The community has pushed the space forward with many creative ideas and powerful models. It’s exciting to see how fast things are moving — even though many fundamental challenges remain. One of the biggest is that cells are not static. They move through trajectories during development, immune responses, and disease. Yet most current models still predict perturbation effects within a single state, rather than how early perturbations propagate across future states and reshape downstream outcomes. To address this, we developed PerturbGen, a trajectory-aware generative AI model that predicts how genetic perturbations reshape downstream cellular states. Huge credit to the people who made this work possible. Thanks to co-first authors Kevin Ly, Adib Miraki, Tomoya Isobe, AmirHoss3in Vahidi, Delshad Vaghari & Anthony Rostron. Special recognition to Kevin Ly and Adib Miraki for driving this work over the finish line. Grateful for our outstanding collaborators from Haniffa Lab, Bertie Gottgens lab Gosia Trynka and many others — a true cross-institute effort across Cambridge Stem Cell Institute, Open Targets ,Wellcome Sanger Institute and Cambridge University.🎉 PerturbGen learns transcriptional dynamics across cellular trajectories. By introducing perturbations at an early source state, it can simulate how these effects propagate into future states along differentiation trajectories. Scaling this across genes enables the creation of dynamic in silico perturbation atlases — maps of how perturbations reshape biological trajectories over time. We explored this idea across three biological questions. First, in a human in vivo LPS immune challenge, PerturbGen predicted that perturbing a transient IL1B signal dampens downstream inflammatory programs in myeloid cells, with pathway changes reversing signatures observed in an independent IL-1β stimulation experiment. Second, in human hematopoiesis, PerturbGen predicted transcriptional responses to CRISPR transcription factor knockouts and enabled construction of perturbation atlases revealing lineage- and age-specific regulatory programs. These programs could also be linked to human genetics and blood diseases, including recapitulation of signatures associated with ETV6-related thrombocytopenia. Finally, we asked whether perturbation modeling could help improve complex tissue models. We built a dynamic perturbation atlas of human skin organoids to identify perturbations that could guideorganoid cells towardhuman fetal skin states. PerturbGen prioritized activation of Wnt signaling via GSK3β inhibition. Experimental validation confirmed the prediction: treatment with CHIR99021 induced stromal gene programs and shifted organoid fibroblasts toward transcriptional states observed in fetal skin stroma. Together, these results show how trajectory-aware perturbation modeling can connect gene perturbations to developmental programs, human genetics, disease mechanisms, and experimental interventions. More broadly, we think these point toward a future where single-cell atlases become predictive systems. As atlases expand across tissues, developmental windows, and modalities, models like PerturbGen could enable dynamic, virtual perturbation atlases— allowing us to simulate interventions, generate hypotheses, and design experiments before stepping into the lab. Preprint Code Excited to see how the community builds on this work.
2:46

Excited to share our new work. Over the past decade, single-cell genomics has transformed our ability to map cellular systems. But a major question remains: Can we predict how perturbations reshape cellular trajectories over time? In 2018, we first showed that it is possible to predict cellular responses to perturbations — ranging from disease signals to chemical treatments — even in unseen contexts. In 2022, we introduced CPA (MSB 2022; NeurIPS 2022), extending this idea to predict responses to unseen chemical and genetic perturbations, including their combinations. Since then, the field of perturbation modeling has grown enormously. The community has pushed the space forward with many creative ideas and powerful models. It’s exciting to see how fast things are moving — even though many fundamental challenges remain. One of the biggest is that cells are not static. They move through trajectories during development, immune responses, and disease. Yet most current models still predict perturbation effects within a single state, rather than how early perturbations propagate across future states and reshape downstream outcomes. To address this, we developed PerturbGen, a trajectory-aware generative AI model that predicts how genetic perturbations reshape downstream cellular states. Huge credit to the people who made this work possible. Thanks to co-first authors Kevin Ly, Adib Miraki, Tomoya Isobe, AmirHoss3in Vahidi, Delshad Vaghari & Anthony Rostron. Special recognition to Kevin Ly and Adib Miraki for driving this work over the finish line. Grateful for our outstanding collaborators from Haniffa Lab, Bertie Gottgens lab Gosia Trynka and many others — a true cross-institute effort across Cambridge Stem Cell Institute, Open Targets ,Wellcome Sanger Institute and Cambridge University.🎉 PerturbGen learns transcriptional dynamics across cellular trajectories. By introducing perturbations at an early source state, it can simulate how these effects propagate into future states along differentiation trajectories. Scaling this across genes enables the creation of dynamic in silico perturbation atlases — maps of how perturbations reshape biological trajectories over time. We explored this idea across three biological questions. First, in a human in vivo LPS immune challenge, PerturbGen predicted that perturbing a transient IL1B signal dampens downstream inflammatory programs in myeloid cells, with pathway changes reversing signatures observed in an independent IL-1β stimulation experiment. Second, in human hematopoiesis, PerturbGen predicted transcriptional responses to CRISPR transcription factor knockouts and enabled construction of perturbation atlases revealing lineage- and age-specific regulatory programs. These programs could also be linked to human genetics and blood diseases, including recapitulation of signatures associated with ETV6-related thrombocytopenia. Finally, we asked whether perturbation modeling could help improve complex tissue models. We built a dynamic perturbation atlas of human skin organoids to identify perturbations that could guideorganoid cells towardhuman fetal skin states. PerturbGen prioritized activation of Wnt signaling via GSK3β inhibition. Experimental validation confirmed the prediction: treatment with CHIR99021 induced stromal gene programs and shifted organoid fibroblasts toward transcriptional states observed in fetal skin stroma. Together, these results show how trajectory-aware perturbation modeling can connect gene perturbations to developmental programs, human genetics, disease mechanisms, and experimental interventions. More broadly, we think these point toward a future where single-cell atlases become predictive systems. As atlases expand across tissues, developmental windows, and modalities, models like PerturbGen could enable dynamic, virtual perturbation atlases— allowing us to simulate interventions, generate hypotheses, and design experiments before stepping into the lab. Preprint Code Excited to see how the community builds on this work.

Mo Lotfollahi

17,230 views • 8 months ago

Evolution says you can turn anything into anything if you tweak it long enough. But that house of cards collapses immediately when you realize what it takes to make even a single cell. They say new cell types arise one mutation at a time. Is that plausible? Here is the issue: Different cell types are created through a massive coordinated stack of genetic information, and it's not all just about DNA sequences. First you need a whole suite of new genetic systems related to the cell: new genes for the cell’s internal systems, instructions for its specific functional role in the body, the regulators that coordinate everything, markers to tell them where to go and when, and the controls that keep the wrong programs off. But the DNA structure itself must also be manipulated. You see, every cell in your body contains the DNA instructions to make any other cell. So how does a bone cell become a bone cell instead of a skin cell? Because of how the DNA is folded. DNA is folded in certain ways for different cell types, so that only the "bone cell instructions" are able to be read by the "cell construction program." The DNA fold decides which programs are open for each cell type. This is critical. If DNA is not folded properly so that specific cell-type information can be read, catastrophe ensues. Then you need placement blueprints, so the cells form in the right place, in the right number, and lock together to form a functional tissue. These developmental instructions are passed from parent to offspring - but research shows that tweaking these master developmental programs ends badly for the child. One mutation at a time cannot invent that entire programming stack and keep it coordinated. Research has confirmed this. To convert one cell type into another in the lab, researchers have to force many changes to happen at the same time in coordination. And even when directed by the researcher, cell conversion almost always fails. Change the sequence without the fold and you cannot get a new cell type. Fold the DNA without the new systems and you can only rearrange old programs. Make the cell without the map and it shows up in the wrong place. Break one layer in the stack and the others do not patiently wait - they fail. So creating a new cell type is not simply "tweak a gene one at a time." It requires a coordinated stack of entirely new programmed instructions: > New suite of internal systems, programmed into the DNA > New functional DNA fold so only the right systems are activated > New mapping blueprints so everything goes where it's needed Step-by-step mutation does not coordinate those layers. It produces broken cells before it produces new ones. Coordination of this level of complexity has only ever been known to come from one place: Intelligence.
0:29

Evolution says you can turn anything into anything if you tweak it long enough. But that house of cards collapses immediately when you realize what it takes to make even a single cell. They say new cell types arise one mutation at a time. Is that plausible? Here is the issue: Different cell types are created through a massive coordinated stack of genetic information, and it's not all just about DNA sequences. First you need a whole suite of new genetic systems related to the cell: new genes for the cell’s internal systems, instructions for its specific functional role in the body, the regulators that coordinate everything, markers to tell them where to go and when, and the controls that keep the wrong programs off. But the DNA structure itself must also be manipulated. You see, every cell in your body contains the DNA instructions to make any other cell. So how does a bone cell become a bone cell instead of a skin cell? Because of how the DNA is folded. DNA is folded in certain ways for different cell types, so that only the "bone cell instructions" are able to be read by the "cell construction program." The DNA fold decides which programs are open for each cell type. This is critical. If DNA is not folded properly so that specific cell-type information can be read, catastrophe ensues. Then you need placement blueprints, so the cells form in the right place, in the right number, and lock together to form a functional tissue. These developmental instructions are passed from parent to offspring - but research shows that tweaking these master developmental programs ends badly for the child. One mutation at a time cannot invent that entire programming stack and keep it coordinated. Research has confirmed this. To convert one cell type into another in the lab, researchers have to force many changes to happen at the same time in coordination. And even when directed by the researcher, cell conversion almost always fails. Change the sequence without the fold and you cannot get a new cell type. Fold the DNA without the new systems and you can only rearrange old programs. Make the cell without the map and it shows up in the wrong place. Break one layer in the stack and the others do not patiently wait - they fail. So creating a new cell type is not simply "tweak a gene one at a time." It requires a coordinated stack of entirely new programmed instructions: > New suite of internal systems, programmed into the DNA > New functional DNA fold so only the right systems are activated > New mapping blueprints so everything goes where it's needed Step-by-step mutation does not coordinate those layers. It produces broken cells before it produces new ones. Coordination of this level of complexity has only ever been known to come from one place: Intelligence.

Divinely Designed

23,703 views • 2 months ago

Scientists just figured out how to reverse aging using AI. And this is a massive breakthrough. We can now reprogram any human cell back to age 20. Heart cells, brain cells, skin cells, all reset to their biological prime. And here’s the wildest part…the technology to do this, has already existed since 2012 (it won the Nobel Prize). But the real breakthrough wasn’t possible until this year, when they supercharged it with AI. It’s a wild story. So in 2006, scientists discovered Yamanaka factors. They’re proteins that can basically convert any normal cell into a universal stem cell. Now this was a huge deal, because these stem cells are basically like magic healers. If you have torn muscle tissue, you could inject these stem cells into the area and they will turn into the youthful muscle cells you need. So Yamanaka factors were this insane breakthrough, because they allowed any human to turn any cell you already have into these magic healers. But, there was one big problem… It turns out, the original Yamanaka factors weren’t very good at this stem cell conversion. They could do it, but they just weren’t very reliable. Enter OpenAI...and this is where things get crazy. OpenAI designed a special AI model built specifically to create new proteins. Think of it like ChatGPT but for protein engineering. So they took all the Yamanaka research and asked this new AI to go ham on improving it. And get this… Their version was 50x more effective than the original. They tested it on 50 year old cells and it successfully started repairing 30% of their cells in just 7 days. This is just science fiction…it actually happened. And it sounds crazy, but in a few years, humans will be able to take a shot that will literally reverse the age of their cells.
1:32

Scientists just figured out how to reverse aging using AI. And this is a massive breakthrough. We can now reprogram any human cell back to age 20. Heart cells, brain cells, skin cells, all reset to their biological prime. And here’s the wildest part…the technology to do this, has already existed since 2012 (it won the Nobel Prize). But the real breakthrough wasn’t possible until this year, when they supercharged it with AI. It’s a wild story. So in 2006, scientists discovered Yamanaka factors. They’re proteins that can basically convert any normal cell into a universal stem cell. Now this was a huge deal, because these stem cells are basically like magic healers. If you have torn muscle tissue, you could inject these stem cells into the area and they will turn into the youthful muscle cells you need. So Yamanaka factors were this insane breakthrough, because they allowed any human to turn any cell you already have into these magic healers. But, there was one big problem… It turns out, the original Yamanaka factors weren’t very good at this stem cell conversion. They could do it, but they just weren’t very reliable. Enter OpenAI...and this is where things get crazy. OpenAI designed a special AI model built specifically to create new proteins. Think of it like ChatGPT but for protein engineering. So they took all the Yamanaka research and asked this new AI to go ham on improving it. And get this… Their version was 50x more effective than the original. They tested it on 50 year old cells and it successfully started repairing 30% of their cells in just 7 days. This is just science fiction…it actually happened. And it sounds crazy, but in a few years, humans will be able to take a shot that will literally reverse the age of their cells.

Whiplash347

68,643 views • 1 year ago

How does an embryo reliably "compute" its form - "cell by cell" - using only local interactions and mechanics, yet produce a precise global body plan? I’m excited to share our Nature Methods paper "MultiCell: geometric learning in multicellular development", presenting #AIxBiology research led by Haiqian Yang and the result of a great collaboration with Ming Guo, George Roy, Tomer Stern, Anh Nguyen and Dapeng Bi. A long-standing challenge in developmental biology is to predict how thousands of cells collectively self-organize as tissues fold, divide, and rearrange. In MultiCell, we represent a developing embryo as a dual graph that unifies two complementary views of tissue mechanics with single-cell resolution: cells as moving points (granular) and cells as a connected foam (junction network). This lets the model learn dynamics from both geometry and cell–cell connectivity. On whole-embryo 4D light-sheet movies of Drosophila gastrulation (~5,000 cells), our model predicts key cell behaviors and the timing of events, including junction loss, rearrangements, and divisions with high accuracy, at single-cell resolution. Beyond prediction, the same representation supports robust time alignment across embryos and offers interpretable activation maps that highlight the morphogenetic "drivers" of development. The broader goal is a foundation for cell-by-cell forecasting in more complex tissues, and eventually for detecting subtle dynamical signatures of disease. Kudos to the team for this inspiring collaboration with brilliant researchers to push the boundary of AI for biology! Citation: Yang, H., Roy, G., Nguyen, A.Q., Buehler, M.J., et al. MultiCell: geometric learning in multicellular development. Nature Methods (2025), DOI: 10.1038/s41592-025-02983-x Code/data links are in the manuscript.
0:46

How does an embryo reliably "compute" its form - "cell by cell" - using only local interactions and mechanics, yet produce a precise global body plan? I’m excited to share our Nature Methods paper "MultiCell: geometric learning in multicellular development", presenting #AIxBiology research led by Haiqian Yang and the result of a great collaboration with Ming Guo, George Roy, Tomer Stern, Anh Nguyen and Dapeng Bi. A long-standing challenge in developmental biology is to predict how thousands of cells collectively self-organize as tissues fold, divide, and rearrange. In MultiCell, we represent a developing embryo as a dual graph that unifies two complementary views of tissue mechanics with single-cell resolution: cells as moving points (granular) and cells as a connected foam (junction network). This lets the model learn dynamics from both geometry and cell–cell connectivity. On whole-embryo 4D light-sheet movies of Drosophila gastrulation (~5,000 cells), our model predicts key cell behaviors and the timing of events, including junction loss, rearrangements, and divisions with high accuracy, at single-cell resolution. Beyond prediction, the same representation supports robust time alignment across embryos and offers interpretable activation maps that highlight the morphogenetic "drivers" of development. The broader goal is a foundation for cell-by-cell forecasting in more complex tissues, and eventually for detecting subtle dynamical signatures of disease. Kudos to the team for this inspiring collaboration with brilliant researchers to push the boundary of AI for biology! Citation: Yang, H., Roy, G., Nguyen, A.Q., Buehler, M.J., et al. MultiCell: geometric learning in multicellular development. Nature Methods (2025), DOI: 10.1038/s41592-025-02983-x Code/data links are in the manuscript.

Markus J. Buehler

388,551 views • 10 months ago

We are thrilled to share that our first paper from my new lab, Spateo ( for spatiotemporal modeling of molecular holograms, is now online in Cell: Spateo is a comprehensive analytical framework for 3D whole-embryo spatiotemporal modeling. Its advanced features include: • 3D alignment and reconstruction at the whole-mouse-embryo scale (see the animation). • 3D spatial domain digitization and cell-cell communication analysis to understand spatial gene expression gradients and both inter- and intracellular communication. • 3D morphometric and volumetric analyses along with 3D morphogenesis vector field modeling to quantify dynamics such as surface area, volume, and cell density across organs, and to dissect the interplay between morphogenesis factors and cell migration. • A “Google Earth”-like browser, Spateo-viewer ( and for interactive and intuitive exploration of 3D spatial data. • Additional features, such as RNA signal-based single-cell segmentation. We are also honored that Nature “News and Views” has highlighted this work as well: This is really an amazing outcome after two years' heroic revision process that rewrite the entire paper using a new data ( for whole mouse embryos.
1:24

We are thrilled to share that our first paper from my new lab, Spateo ( for spatiotemporal modeling of molecular holograms, is now online in Cell: Spateo is a comprehensive analytical framework for 3D whole-embryo spatiotemporal modeling. Its advanced features include: • 3D alignment and reconstruction at the whole-mouse-embryo scale (see the animation). • 3D spatial domain digitization and cell-cell communication analysis to understand spatial gene expression gradients and both inter- and intracellular communication. • 3D morphometric and volumetric analyses along with 3D morphogenesis vector field modeling to quantify dynamics such as surface area, volume, and cell density across organs, and to dissect the interplay between morphogenesis factors and cell migration. • A “Google Earth”-like browser, Spateo-viewer ( and for interactive and intuitive exploration of 3D spatial data. • Additional features, such as RNA signal-based single-cell segmentation. We are also honored that Nature “News and Views” has highlighted this work as well: This is really an amazing outcome after two years' heroic revision process that rewrite the entire paper using a new data ( for whole mouse embryos.

evo-devo

58,572 views • 2 years ago

A whole new wave of disease therapies have started to arrive. They're based on lab made stem cells that can reprogram the body. Folks like Jeff Bezos, Sam Altman and Brian Armstrong have invested billions of dollars into this technology. @nabsicle is trying to mass produce these new stem cells to make the therapies cheaper and more widespread. This is our full episode with her where we get deep into iPSCs and how this technology works. The Core Memory podcast is on all major platforms and on YouTube down below. Brex and @e1ventures are very smart and help us bring the world's smartest podcast to you. Enjoy! Timestamps 00:00 Intro 01:22 Cell Rejuvenation 02:30 AI and Biotech 06:18 Lab-Made vs Natural Stem Cells 08:25 Programming Cell States 10:07 Reversing Parkinson's 12:15 Japan’s Stem Cell Trials 16:46 Personalized Cells 21:19 Cellino’s Laser Bubble Technology 31:34 Stem Cell Fabs 38:33 Personalized Medicine 43:02 Lowering Stem Cell Therapy Costs 46:31 Can Humans Live Forever?
1:17:51

A whole new wave of disease therapies have started to arrive. They're based on lab made stem cells that can reprogram the body. Folks like Jeff Bezos, Sam Altman and Brian Armstrong have invested billions of dollars into this technology. @nabsicle is trying to mass produce these new stem cells to make the therapies cheaper and more widespread. This is our full episode with her where we get deep into iPSCs and how this technology works. The Core Memory podcast is on all major platforms and on YouTube down below. Brex and @e1ventures are very smart and help us bring the world's smartest podcast to you. Enjoy! Timestamps 00:00 Intro 01:22 Cell Rejuvenation 02:30 AI and Biotech 06:18 Lab-Made vs Natural Stem Cells 08:25 Programming Cell States 10:07 Reversing Parkinson's 12:15 Japan’s Stem Cell Trials 16:46 Personalized Cells 21:19 Cellino’s Laser Bubble Technology 31:34 Stem Cell Fabs 38:33 Personalized Medicine 43:02 Lowering Stem Cell Therapy Costs 46:31 Can Humans Live Forever?

Ashlee Vance

21,622 views • 8 months ago

I'm creating 3D tours of open single cell RNA-seq data in UMAP space. Here, using data from 50k brain cells by the allen institute . ( The UMAP position and class of every cell is read into Blender, realized as a stylized, class-specific model I sculpted
1:06

I'm creating 3D tours of open single cell RNA-seq data in UMAP space. Here, using data from 50k brain cells by the allen institute . ( The UMAP position and class of every cell is read into Blender, realized as a stylized, class-specific model I sculpted

Brad Krajina

48,156 views • 3 years ago

🚨Here's what a lot of people misunderstand about cancer treatment, says drpaulmarik: "Cancer is not homogeneous. The somatic mutation theory—which is the current theory in which treatment is based—posits that you have a mutation in a single cell, and that gives rise to a whole population of cells that look the same and have the same mutation. But the Cancer Genome Atlas has shown that that theory is completely wrong. The cancer cells are very heterogeneous, so they're made up of very different populations of cells with different mutations, and one of the populations is the cancer stem cell. It's a sub-population of the cancer. These are generally slow-growing, but they're distinct in that they have the ability to divide indefinitely and grow indefinitely, and can change their characteristics. Basically, if you get rid of the fast-dividing cells, which is the cancer, you're left with the stem cells, which then become the roots, which grow back to form the tumor" sometimes years later. Conventional chemotherapy gets rid of the fast-dividing regular cancer cells but *NOT* the stem cells. So the key question is: how do you get rid of the stem cells? “There are a number of repurposed drugs that do it, and this has been well-established in scientific medical literature. One of the most effective treatments to knock out the stem cell is the famous horse deworming medicine," says drpaulmarik. Yes, ivermectin. Independent Medical Alliance
6:30

🚨Here's what a lot of people misunderstand about cancer treatment, says drpaulmarik: "Cancer is not homogeneous. The somatic mutation theory—which is the current theory in which treatment is based—posits that you have a mutation in a single cell, and that gives rise to a whole population of cells that look the same and have the same mutation. But the Cancer Genome Atlas has shown that that theory is completely wrong. The cancer cells are very heterogeneous, so they're made up of very different populations of cells with different mutations, and one of the populations is the cancer stem cell. It's a sub-population of the cancer. These are generally slow-growing, but they're distinct in that they have the ability to divide indefinitely and grow indefinitely, and can change their characteristics. Basically, if you get rid of the fast-dividing cells, which is the cancer, you're left with the stem cells, which then become the roots, which grow back to form the tumor" sometimes years later. Conventional chemotherapy gets rid of the fast-dividing regular cancer cells but *NOT* the stem cells. So the key question is: how do you get rid of the stem cells? “There are a number of repurposed drugs that do it, and this has been well-established in scientific medical literature. One of the most effective treatments to knock out the stem cell is the famous horse deworming medicine," says drpaulmarik. Yes, ivermectin. Independent Medical Alliance

Jan Jekielek

96,002 views • 1 year ago

1/n: 🔬We introduce NicheCompass (NC) a graph learning model to build spatial atlases across millions of cells, quantitatively characterize cell niches, infer cellular communication, and map new data into these atlases. NC can be used with various spatial technologies, ranging from cellular to subcellular resolution including even multimodal approaches. We have explored interesting biology by analyzing cell niches spanning from development to cancer. Led by Sebastian Birk in a fun colab with @cartalop: paper:
1:04

1/n: 🔬We introduce NicheCompass (NC) a graph learning model to build spatial atlases across millions of cells, quantitatively characterize cell niches, infer cellular communication, and map new data into these atlases. NC can be used with various spatial technologies, ranging from cellular to subcellular resolution including even multimodal approaches. We have explored interesting biology by analyzing cell niches spanning from development to cancer. Led by Sebastian Birk in a fun colab with @cartalop: paper:

Mo Lotfollahi

61,805 views • 2 years ago

Excited to share TERRA, a tissue world model 🧬 Over ~1.5 years we ran a large data-generation + modelling effort to build a world model for human tissues, pretrained on 112M cells from spatial transcriptomics (mostly Xenium 5000-plex + public data). It's built on one of the largest human spatial transcriptomics corpora assembled to date, spanning 20 tissues across development, health and 26 disease conditions, ~two-thirds newly generated in-house. Why a "world model" for tissue? Images have universal representations (ViT/DINOv3), so do proteins (ESM, Alex Rives) and pathology (UNI, Faisal Mahmood). We've worked hard to build something similar for human tissue: one model that captures its multi-scale logic, genes → cells → their native microenvironments. Like the JEPA approach Yann LeCun has championed, TERRA learns by prediction in embedding space, but for human tissue. How it works: it tokenises each cell together with its nearest neighbours into one sequence while keeping every gene's identity, then masks part of a neighbourhood and predicts the representation of the hidden part, not raw noisy counts. From one backbone it reads out three scales, gene embeddings (what a gene is doing in a cell and its niche), cell embeddings (cell type and state) and neighbourhood embeddings (the niche), and because it keeps gene-level resolution it can knock a gene out in silico and predict the response. Applied entirely zero-shot, TERRA maps and perturbs human tissue across unseen organs, diseases and technologies, outperforming existing spatial approaches. Three take-homes: 1️⃣ One model, any tissue. A single pretrained backbone provides tissue representations zero-shot, handling genes, cells and niches across organs and platforms, off the shelf. 2️⃣ New biology, development to clinic. We built a new spatial atlas of the developing human pancreas and found an islet-associated capillary state that looks like a precursor of mature islet vasculature. In kidney, TERRA's in silico knockouts predicted the tissue-injury programme from cancer immunotherapy (checkpoint blockade), confirmed in treated kidneys, detected in blood, and linked to declining kidney function. 3️⃣ A grammar of tissue architecture. By coupling each cell's state to its niche, TERRA defines recurring cross-organ "archetypes" of macrophage neighbourhoods, including a tumour-boundary niche that tracks poor survival in kidney cancer. TERRA is already in use: it powered our recent skin atlas of hidden immune-memory niches ( with more studies coming soon. This was an amazing collaboration between clinicians, machine-learning scientists and cell biologists 🙏 Led by Sebastian Birk, Vali Sanian Amirhosein Vahidi, Samuel Ogden, Daniyal Jafree, Adib Miraki, Carlo Leonardi and Arpit Merchant, with Lassi Paavolainen, Menna Clatworthy, Omer Ali Bayraktar, Muzlifah Haniffa, Tom Mitchell and Mostafa Bakhti. Huge thanks too to everyone who shared data and helped along the way. What excites me most is seeing how the community builds on this. The model, code and tutorials are all public, so anyone can run TERRA on their own tissues, extend it, or build new models on top. Huge thanks to the whole team across Wellcome Sanger Institute and our many collaborators. 📄 Paper: 💻 Code: 🤗 Model: #SpatialTranscriptomics #SpatialGenomics #FoundationModels #AI4Science #MachineLearning #ComputationalBiology #SingleCell #WorldModels
2:15

Excited to share TERRA, a tissue world model 🧬 Over ~1.5 years we ran a large data-generation + modelling effort to build a world model for human tissues, pretrained on 112M cells from spatial transcriptomics (mostly Xenium 5000-plex + public data). It's built on one of the largest human spatial transcriptomics corpora assembled to date, spanning 20 tissues across development, health and 26 disease conditions, ~two-thirds newly generated in-house. Why a "world model" for tissue? Images have universal representations (ViT/DINOv3), so do proteins (ESM, Alex Rives) and pathology (UNI, Faisal Mahmood). We've worked hard to build something similar for human tissue: one model that captures its multi-scale logic, genes → cells → their native microenvironments. Like the JEPA approach Yann LeCun has championed, TERRA learns by prediction in embedding space, but for human tissue. How it works: it tokenises each cell together with its nearest neighbours into one sequence while keeping every gene's identity, then masks part of a neighbourhood and predicts the representation of the hidden part, not raw noisy counts. From one backbone it reads out three scales, gene embeddings (what a gene is doing in a cell and its niche), cell embeddings (cell type and state) and neighbourhood embeddings (the niche), and because it keeps gene-level resolution it can knock a gene out in silico and predict the response. Applied entirely zero-shot, TERRA maps and perturbs human tissue across unseen organs, diseases and technologies, outperforming existing spatial approaches. Three take-homes: 1️⃣ One model, any tissue. A single pretrained backbone provides tissue representations zero-shot, handling genes, cells and niches across organs and platforms, off the shelf. 2️⃣ New biology, development to clinic. We built a new spatial atlas of the developing human pancreas and found an islet-associated capillary state that looks like a precursor of mature islet vasculature. In kidney, TERRA's in silico knockouts predicted the tissue-injury programme from cancer immunotherapy (checkpoint blockade), confirmed in treated kidneys, detected in blood, and linked to declining kidney function. 3️⃣ A grammar of tissue architecture. By coupling each cell's state to its niche, TERRA defines recurring cross-organ "archetypes" of macrophage neighbourhoods, including a tumour-boundary niche that tracks poor survival in kidney cancer. TERRA is already in use: it powered our recent skin atlas of hidden immune-memory niches ( with more studies coming soon. This was an amazing collaboration between clinicians, machine-learning scientists and cell biologists 🙏 Led by Sebastian Birk, Vali Sanian Amirhosein Vahidi, Samuel Ogden, Daniyal Jafree, Adib Miraki, Carlo Leonardi and Arpit Merchant, with Lassi Paavolainen, Menna Clatworthy, Omer Ali Bayraktar, Muzlifah Haniffa, Tom Mitchell and Mostafa Bakhti. Huge thanks too to everyone who shared data and helped along the way. What excites me most is seeing how the community builds on this. The model, code and tutorials are all public, so anyone can run TERRA on their own tissues, extend it, or build new models on top. Huge thanks to the whole team across Wellcome Sanger Institute and our many collaborators. 📄 Paper: 💻 Code: 🤗 Model: #SpatialTranscriptomics #SpatialGenomics #FoundationModels #AI4Science #MachineLearning #ComputationalBiology #SingleCell #WorldModels

Mo Lotfollahi

51,634 views • 3 months ago

The Human Cell Atlas, a global effort to map and profile the body’s cells, will advance our understanding of the human body in ways not previously possible. How? Muzlifah Haniffa from the Wellcome Sanger Institute tells us more about this ambitious project we're proud to fund 📽️
0:52

The Human Cell Atlas, a global effort to map and profile the body’s cells, will advance our understanding of the human body in ways not previously possible. How? Muzlifah Haniffa from the Wellcome Sanger Institute tells us more about this ambitious project we're proud to fund 📽️

Wellcome

10,158 views • 3 years ago

CHOIR is now online Nature Genetics! CHOIR is a new clustering method for single-cell data that evaluates whether clusters represent statistically distinct cell populations. CHOIR scales to millions of cells and works with single-/multi-omic data of any type!
1:42

CHOIR is now online Nature Genetics! CHOIR is a new clustering method for single-cell data that evaluates whether clusters represent statistically distinct cell populations. CHOIR scales to millions of cells and works with single-/multi-omic data of any type!

Cathrine Sant

55,466 views • 1 year ago

A few months ago, researchers made a synthetic cell that grows, eats, and divides. But it only grows for about five generations, and then it “breaks down” completely and stops moving. This is the reason it’s not yet considered “alive.” When I asked Kate Adamala, the lead author behind SpudCell, why she thinks this happens, she said something really simple and somewhat surprising: The cell cannot break down its own waste. In other words, it has all these proteins and RNA and DNA, but has no way to break them down and reuse the building blocks when they inevitably fail. So when a ribosome stops working, it just sits inside the cell and remains there for many generations. With each generation, more of the cell is filled up with broken pieces. Now, you could add a protease to these cells, but proteases are blunt-force instruments. They would destroy both working proteins *and* the broken ones, and so probably kill the cell. And when I asked Kate why we can't just use the enzymes and chaperones that natural cells use, she said there are too many of them. If you try to recreate all of biology's quality control processes, you end up with a hugely bloated metabolism inside what is supposed to be a minimal cell. The issue, then — the fundamental bottleneck — is that we do not have good tools to recognize WHAT is broken inside a cell and *then* destroy them while leaving healthy variants unscathed. My hunch here is that we need better molecular sensors. If we could distinguish between different forms of a protein, without destroying a cell entirely, then perhaps we could couple that recognition to a selective protease and recycle only the things that are broken. We also just don’t understand, at a high enough level, what happens when a ribosome breaks, and thus we can’t design good probes to sense those broken molecules. In trying to create a synthetic cell, then, we might also arrive at ideas for much better tools. Trying to create life is a really good way to figure out all the obstacles and stuff we don’t yet understand. But when Kate says she thinks they'll achieve perpetual cell division by 2030 (in the interview), I'm kind of skeptical. There's a lot of work to do here, and the tools we tend to use as biologists are often much broader than the precise, highly-evolved systems we find in nature.
2:55

A few months ago, researchers made a synthetic cell that grows, eats, and divides. But it only grows for about five generations, and then it “breaks down” completely and stops moving. This is the reason it’s not yet considered “alive.” When I asked Kate Adamala, the lead author behind SpudCell, why she thinks this happens, she said something really simple and somewhat surprising: The cell cannot break down its own waste. In other words, it has all these proteins and RNA and DNA, but has no way to break them down and reuse the building blocks when they inevitably fail. So when a ribosome stops working, it just sits inside the cell and remains there for many generations. With each generation, more of the cell is filled up with broken pieces. Now, you could add a protease to these cells, but proteases are blunt-force instruments. They would destroy both working proteins *and* the broken ones, and so probably kill the cell. And when I asked Kate why we can't just use the enzymes and chaperones that natural cells use, she said there are too many of them. If you try to recreate all of biology's quality control processes, you end up with a hugely bloated metabolism inside what is supposed to be a minimal cell. The issue, then — the fundamental bottleneck — is that we do not have good tools to recognize WHAT is broken inside a cell and *then* destroy them while leaving healthy variants unscathed. My hunch here is that we need better molecular sensors. If we could distinguish between different forms of a protein, without destroying a cell entirely, then perhaps we could couple that recognition to a selective protease and recycle only the things that are broken. We also just don’t understand, at a high enough level, what happens when a ribosome breaks, and thus we can’t design good probes to sense those broken molecules. In trying to create a synthetic cell, then, we might also arrive at ideas for much better tools. Trying to create life is a really good way to figure out all the obstacles and stuff we don’t yet understand. But when Kate says she thinks they'll achieve perpetual cell division by 2030 (in the interview), I'm kind of skeptical. There's a lot of work to do here, and the tools we tend to use as biologists are often much broader than the precise, highly-evolved systems we find in nature.

Niko McCarty.

11,388 views • 1 month ago

Will we ever be able to simulate a living cell? In theory. Molecules collide, proteins fold, and all of these things can be modeled on a computer. But there are so many unknowns, and so much compute would be required, that a mechanistic model of the cell remains a distant dream. Still, many research groups are trying to build cells "from the bottom up," mostly by stringing together mathematical equations that represent different parts of the cell. By *trying* to build an accurate model of the cell, they hope to improve our own understanding of how biology works. In doing so, they also maintain legibility, meaning that humans can understand and interpret all the equations used to construct the model. But Adam Green argues that legibility is a constraint on our models. He thinks that "human legibility, our ability to understand a system...has been limiting us." To truly accelerate biomedical progress, Green thinks that we should discard assumptions about legibility in favor of "more black box" models. Neural networks can make sense of biological data in a way that hard-coded equations cannot. This is not a new argument. A 2019 essay by Bert Hubert, called "Is biology too complex to ever understand?" makes much the same point. In that essay, Hubert writes: "There is no rule that says nature cannot be more complex than our brains can handle." Instead of trying to make sense of biology via reductionist observations and mechanisms, Hubert argues that we should just gather everything into databases "that might enable computers to make sense of what we have learned." (This vision is now playing out in many research groups.) This "black box" approach might be useful (in terms of modeling cells to cure diseases, say) but would be unsatisfying in the sense that it doesn't deepen our own understanding of the universe. And isn't that a key part of science? Is science for humans -- an act that satisfies some itch -- or rather a means to an end? I think both approaches are useful. The black box models may help us cure diseases faster by making useful predictions that our brains cannot (currently) comprehend, but the mechanistic models deepen our own understanding of how cells work.
2:26

Will we ever be able to simulate a living cell? In theory. Molecules collide, proteins fold, and all of these things can be modeled on a computer. But there are so many unknowns, and so much compute would be required, that a mechanistic model of the cell remains a distant dream. Still, many research groups are trying to build cells "from the bottom up," mostly by stringing together mathematical equations that represent different parts of the cell. By *trying* to build an accurate model of the cell, they hope to improve our own understanding of how biology works. In doing so, they also maintain legibility, meaning that humans can understand and interpret all the equations used to construct the model. But Adam Green argues that legibility is a constraint on our models. He thinks that "human legibility, our ability to understand a system...has been limiting us." To truly accelerate biomedical progress, Green thinks that we should discard assumptions about legibility in favor of "more black box" models. Neural networks can make sense of biological data in a way that hard-coded equations cannot. This is not a new argument. A 2019 essay by Bert Hubert, called "Is biology too complex to ever understand?" makes much the same point. In that essay, Hubert writes: "There is no rule that says nature cannot be more complex than our brains can handle." Instead of trying to make sense of biology via reductionist observations and mechanisms, Hubert argues that we should just gather everything into databases "that might enable computers to make sense of what we have learned." (This vision is now playing out in many research groups.) This "black box" approach might be useful (in terms of modeling cells to cure diseases, say) but would be unsatisfying in the sense that it doesn't deepen our own understanding of the universe. And isn't that a key part of science? Is science for humans -- an act that satisfies some itch -- or rather a means to an end? I think both approaches are useful. The black box models may help us cure diseases faster by making useful predictions that our brains cannot (currently) comprehend, but the mechanistic models deepen our own understanding of how cells work.

Niko McCarty.

16,417 views • 4 months ago

Introducing Tahoe-x1 (Tx1) by Tahoe. A 3-billion-parameter, single-cell foundation model that learns unified representations of genes, cells, and drugs, achieving state-of-the-art performance across cancer-relevant cell biology benchmarks, open-sourced on Hugging Face. 🧵
0:31

Introducing Tahoe-x1 (Tx1) by Tahoe. A 3-billion-parameter, single-cell foundation model that learns unified representations of genes, cells, and drugs, achieving state-of-the-art performance across cancer-relevant cell biology benchmarks, open-sourced on Hugging Face. 🧵

Nima Alidoust

172,248 views • 1 year ago

How do we go from a single cell to a complex organism? For 50 yrs experimentally meeting the nutritional demands of a growing organism has been a bottleneck in development. For the first time, we built placental support and predictive tools to make it possible. Watch ↓
4:05

How do we go from a single cell to a complex organism? For 50 yrs experimentally meeting the nutritional demands of a growing organism has been a bottleneck in development. For the first time, we built placental support and predictive tools to make it possible. Watch ↓

Becoming

869,791 views • 10 months ago