date: “2025-07-28” title: “Gauss Law”¶
What do you know?
Great. I’ll gather detailed, connected insights from the past three months on recent academic advances in:
- Single-cell omics (especially spatial omics),
- Computational systems biology,
- Synthetic developmental biology,
…with a focus on work from HHMI labs and other leading groups (like the Whale Lab). The central emphasis will be on developmental circuits, particularly in embryoids, organoids, and in vivo/in vitro models.
I’ll prioritize high-impact publications, tools, software, and databases, and provide a structured, readable synthesis instead of a large list of links.
I’ll let you know when the review is ready.
Recent Advances in Single-Cell Omics, Computational Systems Biology, and Synthetic Developmental Biology (Spring–Summer 2025)¶
Single-Cell Omics: Multi-Omics, Spatial Profiling, and New Modalities¶
Single-cell “omics” technologies continue to rapidly evolve, offering deeper and more integrative views of cellular biology. In the past three months, researchers and companies have unveiled tools that boost multi-omics throughput and extend single-cell analysis to new molecular realms (proteins, metabolites) and spatial contexts:
High-throughput Single-Cell Multi-omics: A new integrated workflow combines single-cell whole-genome and transcriptome analysis in a 384-well format, using automated cell dispensing to process hundreds to thousands of cells in parallel. This approach yields sequencing-ready libraries (DNA + RNA) from each cell in under 10 hours, greatly streamlining multi-omic profiling without laborious cell sorting. Such technology, introduced via a BioSkryb Genomics and Tecan partnership, addresses the growing demand for scalable single-cell multi-omics in research and precision medicine.
Dynamic Single-Cell Metabolomics: Traditional single-cell metabolomics provided only static “snapshots” of metabolite levels. A recent advance enables dynamic metabolomic profiling of individual cells by integrating stable-isotope tracing and high-throughput mass spectrometry. The system introduces labeled substrates (e.g. ^13C-glucose) to cells and uses an automated data pipeline (including an untargeted isotope-tracing analysis platform) to track metabolic fluxes in thousands of single cells. This reveals heterogeneous metabolic activities and pathway usage among cells that would be invisible from bulk measurements. Notably, coupling this platform with a neural-network model allowed researchers to profile co-cultured tumor cells and macrophages, uncovering cell–cell metabolic interactions: for example, they identified distinct tumor-associated macrophage metabolic phenotypes in the mix, consistent with diversity seen in single-cell RNA-seq atlases. Dynamic single-cell metabolomics thus opens a window into how cells influence each other’s metabolism in real time, an important aspect of the tumor microenvironment and other tissues.
Expanding Single-Cell Proteomics & Metabolomics: Single-cell transcriptomics is now routine, but proteomic and metabolomic profiling at single-cell resolution are still emerging. Analytical experts note that while single-cell proteomics is advancing rapidly, single-cell metabolomics remains a major challenge. Due to sensitivity limits, current single-cell mass spectrometry can detect on the order of only tens of metabolites or lipids per cell, versus thousands in bulk samples. Despite this shallow coverage, even limited single-cell metabolomes show striking differences between cells. For instance, in a brain region, certain neurotransmitters or metabolites (serotonin, dopamine, specific amino acids) may be abundant in only a subset of neurons and nearly absent in others. These insights underscore cell-to-cell biochemical heterogeneity that was previously masked. Ongoing improvements in sample preparation and ultrasensitive mass spectrometry (e.g. new ion trapping instruments) are steadily improving single-cell proteome/metabolite depth. Recent product launches (from companies like Bruker and Standard BioTools) and open-source analysis frameworks are focusing on higher throughput and better quantitation in single-cell proteomics, aiming to bring these modalities more on par with single-cell transcriptomics in the near future.
Integrating Spatial Transcriptomics with Single-Cell Data: To connect molecular profiles with tissue architecture, researchers are merging single-cell sequencing with spatial transcriptomics. In July 2025, a team introduced Cellular Mapping of Attributes with Position (CMAP), a computational method to precisely map individual cells (from scRNA-seq) back onto their location in a tissue section. This divide-and-conquer algorithm integrates single-cell transcriptomes with spatial gene expression data, overcoming differences in resolution and coverage between the two sources. Tests on simulated and real datasets showed CMAP can assign cells to exact coordinates, even when there are mismatches or noise between the single-cell and spatial data. By endowing each single cell with a precise spatial address, tools like CMAP allow scientists to dissect fine-grained tissue organization – resolving, for example, spatial patterns of endothelial subtypes or the layout of immune and tumor cells within a tumor microenvironment that might elude standard spot-level spatial analysis. This integration of spatial context with single-cell detail is proving powerful in developmental biology (mapping where each cell type emerges in an embryo) and in pathology (identifying how cancer cells and immune cells are arranged and interact in a tumor).
Computational Systems Biology: AI Models and Theoretical Insights¶
Computational methods are at the heart of recent systems biology breakthroughs, especially in leveraging machine learning to interpret complex biological data and using mathematical modeling to explain emergent behaviors. Over the last quarter, researchers have pushed the envelope with new AI-driven tools and computational models that enhance our ability to predict, manipulate, or understand biological systems:
Causal Generative Models for Single-Cell Data: A July 2025 study introduced CausCell, a framework that combines causal inference with diffusion models (a class of generative AI) to learn explainable representations of single-cell omics data. Unlike black-box machine learning that embeds cells in abstract latent space, CausCell tries to disentangle specific biological “concepts” (e.g. cell cycle stage, disease status, stimulus response) and incorporate known cause–effect relationships among those concepts. By training on single-cell transcriptomic data (including spatial and temporal data) with causal graph constraints, the model can generate counterfactual cell profiles – essentially predicting how a cell’s gene expression would change if one factor could be tweaked. This approach outperformed other representation-learning methods in benchmarks, indicating it more successfully separated underlying factors in the data. The ability to simulate “what-if” scenarios for individual cells (for example, if cell A were infected by a pathogen, how would its state differ?) is an exciting step towards virtual cell models that help probe mechanisms and predict interventions in silico. Such AI-driven virtual cells could eventually allow researchers to test hypotheses on digital twins of cells before verifying in the lab.
Network and Phase-Transition Models of Cell Communities: Systems biologists are also using computational modeling to explain how cell–cell interactions give rise to collective behavior. A notable recent example modeled the metabolic interplay in multicellular systems and discovered a phase transition that might underlie a long-standing puzzle in metabolism. Specifically, the model examined overflow metabolism – the seemingly wasteful excretion of nutrients like lactate by cells even in oxygen-rich conditions (a phenomenon seen in cancer cells and yeast). By integrating spatial metabolic models with statistical physics, researchers showed that when cells exchange metabolites in a diffusion-limited environment, a tipping point is reached: beyond a certain nutrient uptake rate, the system flips from a coordinated state (where cells fully metabolize resources) to an overflow state where metabolic coordination fails. Near this transition, individual cells exhibit heterogeneous metabolic phenotypes – mirroring what single-cell experiments observe in tumors. This theoretical work, published in May 2025, suggests that a breakdown in inter-cellular metabolic cooperation (rather than just cell-intrinsic limits) can drive overflow metabolism, reframing it as an emergent, population-level property. More broadly, it demonstrates how phase transition concepts can illuminate biology, identifying conditions where a multicellular system’s behavior changes qualitatively. Such insights could inform strategies to steer tissue-level outcomes by modulating cell density, nutrient supply, or signaling interactions.
AI-Powered High-Throughput Analysis: Artificial intelligence is increasingly used to handle the deluge of complex data in biology. A very recent example is deepBlastoid, an AI tool developed at KAUST (announced July 22, 2025) to accelerate studies of human embryo models grown in vitro. Labs can produce hundreds of blastoid (blastocyst-like) embryo models, but evaluating their morphology and development is labor-intensive. DeepBlastoid employs deep learning to analyze microscope images of blastoids as accurately as human experts, but ~1000× faster. Trained on over 2,000 annotated images, the system can rapidly grade blastoids and even assess how different chemicals or culture conditions affect their growth, as demonstrated in tests on 10,000 images. This massively increased throughput enables experiments that were previously unfeasible – for example, screening drug effects on early embryonic development or optimizing culture methods in a high-content manner. By automating image-based evaluations of embryo-like structures, AI tools like deepBlastoid free up researchers and allow scaling up studies of early development under many conditions in parallel. It’s a prime example of how computational advances (in computer vision and generative AI) are now accelerating experimental biology.
Using AI to analyze embryo models: Researchers are leveraging deep learning (e.g. the deepBlastoid tool) to analyze lab-grown blastoids – simplified human embryo models – far faster than manual inspection. High-throughput image analysis can evaluate developmental progression and the effects of various compounds on these embryoid structures, aiding both basic research and reproductive medicine. (Image: Microscope view of laboratory-grown embryo model; source: KAUST)*
Synthetic Developmental Biology: Organoids and Embryoids in the Spotlight¶
Synthetic developmental biology – where scientists build or manipulate developing systems to understand how complex organisms form – has seen remarkable progress recently, especially in the realm of organoids (lab-grown miniature organs) and embryoids (embryo-like structures). Over the past few months, advances in growing, controlling, and analyzing these systems have opened new frontiers in developmental biology:
Synthetic Human Embryo Models Reaching New Milestones: Researchers are pushing the boundaries of creating embryo analogues from stem cells. Just over a year ago, in mid-2023, teams in the US, UK, and Israel independently reported the first synthetic human embryo models that mimic certain early developmental stages. These models – often called blastoids or embryoids – were grown from pluripotent stem cells without using sperm or eggs. In the past 3 months, this line of research has continued to mature. The current state-of-the-art embryoid can recapitulate the blastocyst to gastrulation transition: for example, models have been grown that reach the beginning of gastrulation (~day 14 of development), containing cells corresponding to all three embryonic germ layers as well as primordial germ cell precursors (the cells that would form egg/sperm in a natural embryo). Magdalena Żernicka-Goetz, whose lab presented such results, emphasized that these structures are “embryo models,” not fully equivalent to natural embryos, but they strikingly resemble real embryos at that stage. Importantly, they offer a novel window into the “black box” period of human development just after implantation – a phase where many pregnancies fail and which is normally inaccessible for ethical reasons. With these models, scientists can begin to study crucial events like primordial germ cell specification or early axis patterning in vitro. Moving forward, teams (such as Jacob Hanna’s group in Israel) are aiming to extend these embryoids further (e.g. towards the 21-day post-fertilization stage), though significant challenges remain before reaching later organ-forming stages. The recent progress, however, is already raising ethical questions and prompting regulatory bodies (like in the UK) to draft guidelines for research on embryo models that go beyond the traditional 14-day limit. Overall, the past months have solidified synthetic embryology as a burgeoning field, with labs around the world racing to refine human embryo models to study infertility, congenital disorders, and early development in a dish.
Organoids Revealing Developmental Processes: Lab-grown organoids – from brains to blastoids – are not only getting more sophisticated but are being used to dissect developmental mechanisms. A highlight in June 2025 was a long-term live imaging study of human brain organoids that revealed how physical environment cues shape early brain development in vitro. Researchers cultivated unguided cerebral organoids from human iPSCs and used light-sheet microscopy to follow their growth over weeks, tracking single-cell behaviors and even subcellular dynamics in real time. By combining the imaging data with single-cell RNA sequencing, they found that adding an extracellular matrix (ECM) scaffold (like Matrigel) had profound effects on organoid development. The presence of an external matrix promoted the expansion and fusion of internal lumens and encouraged the formation of forebrain-like (telencephalic) tissue, whereas organoids grown without an extrinsic matrix showed abnormal morphology with more neural crest cells and more posterior (caudal) identities. The study pinpointed that mechanosensing pathways (like WNT and Hippo/YAP signaling) were modulated by the matrix: matrix-supported organoids showed spatially restricted expression of WLS (Wnt ligand secretion mediator), marking the earliest emergence of non-telencephalic brain regions, whereas matrix-free organoids had dysregulated patterns. In short, this work provided a “morphodynamic” view of brain development – demonstrating how biomechanical cues (stiffness, physical support) influence cell fate patterning and regionalization in a developing brain. It also offers practical insight: tweaking the culture matrix could improve organoid models of the human brain, yielding more faithful architecture and cell type diversity. More generally, organoids are increasingly used as models of embryonic development. For instance, researchers are growing gastruloids (early embryonic organoids) to study how body axes and germ layers self-organize, and neural tube organoids to examine dorsal-ventral patterning of the spinal cord. These self-organizing systems, observed with cutting-edge imaging and single-cell omics, are shedding light on the coordinate choreography of cell movements and gene expression that underlies embryogenesis.
Synthetic Gene Circuits to Control Development: A key goal of synthetic developmental biology is not only to observe organoids and embryoids, but to steer them – to programmably control cell differentiation and tissue patterning. Recent work has expanded the toolkit for doing exactly this. In June 2025, scientists reported a panel of inducible gene expression systems tailored for use in embryonic stem cells (ESCs). These systems include three that are triggered by small molecules (drugs) and two activated by cell–cell contact signals, all engineered to drive expression of chosen target genes in ESCs with tight control. The researchers systematically tested these synthetic circuits in mouse ESCs and showed that they can be reliably used to modulate developmental pathways – even demonstrating directed differentiation of ESCs into neurons by turning on a proneural factor under these inducible systems. Each control module (for example, a drug-inducible synthetic transcription factor or a contact-activated receptor like SynNotch) has its own dynamics and ideal use cases, but importantly, they can be combined in the same cells to create logical control over developmental decisions. By mixing and matching such tools, one could, say, induce a certain gene only when a cell is in contact with its neighbor and a particular drug is added – achieving spatiotemporal control reminiscent of natural developmental signals. This work, part of a growing effort led by bioengineers, is expanding synthetic biology’s reach into multicellular systems. The payoff will be the ability to construct designer developmental programs: imagine programming stem cells to self-assemble into an organoid with a desired size and cell composition by flipping genetic “switches” at set time points. While still early, the recent demonstration shows these inducible systems function robustly in pluripotent stem cells, encouraging further use to probe how precisely altering gene expression affects pattern formation and cell fate outcomes. Such approaches will help test developmental biology’s theoretical principles by rebuilding circuits in a controllable way (“build to understand”).
Organoids as Platforms for Disease and Therapy: Alongside fundamental advances, the last few months have also seen organoids and embryoids leveraged in more applied directions. For example, patient-derived tumor organoids are being co-cultured with normal organoids (like brain organoids) to study how cancer cells invade and interact with healthy tissue. Multi-omic single-cell analyses of these mixtures can reveal vulnerabilities of cancer cells in a more physiological 3D context. In regenerative medicine, researchers are exploring organoid transplantation and fusion – e.g. merging synthetic embryonic tissues with natural embryos or grafting organoids into models – to bridge the gap between in vitro development and in vivo function. Meanwhile, bioethical and regulatory discussions are intensifying as synthetic embryo models become more realistic. In summary, synthetic developmental biology is at an inflection point: labs can now create increasingly life-like models of early human development and program them to a degree, offering unprecedented insight into how complex life forms emerge from single cells. The past three months have underscored a central theme – by uniting single-cell omics, advanced computation, and synthetic biology, scientists are decoding and even recreating the developmental “blueprints” of life, step by step.
Sources: Recent literature and reports (April–July 2025) were used to compile these findings, including cutting-edge research articles in Nature, Nature Communications, Development, and others, as well as science news releases. These sources provide detailed accounts of the breakthroughs summarized above, reflecting the rapidly evolving state of single-cell analysis, computational modeling, and synthetic developmental systems.
Website¶
I finally got the damn website working right. Before none of the images would render anywhere beside my local computer because of how myst start works. There are two servers. One for the page, one for the static content.
Anyway I figured it out and now my site is just static content. Tried to figure out how to get caddy to serve it with file_server but gave up on that quickly.
I started pulling everything into docker compose. I thought it was sooo easy to set up openwebui, phoenix, and litellm. And it is. Each of them separately outside of docker. Once you start connecting all of them to postgres and need to do things in a particular order and connect the phoenix collectors and all that it becomes a hassle. I’m still about it, but taking it easier than just staying awake another 48 hours resolving everything like I did during this design phase.
immanentize - by me