Codebook Project Findings


The Codebook project has identified the preferred short DNA sequences for 177 understudied proteins that regulate gene function. On August 5, an article about the Codebook project was published in Nature. The authors investigated 332 presumed human transcription factors, which are proteins involved in regulating gene function, and obtained motifs for 177 of them.

A motif is a short set of preferences for DNA letters, showing which sequences a protein binds to more readily. For 130 factors, data from cells showed where such preferences are manifested in the genome. In a 2018 catalog, researchers listed 1,639 presumed human transcription factors, with more than a quarter of them having unknown motifs.

Knowing the motif allows researchers to check if replacing one DNA letter changes the binding of a specific protein and then search for its consequences for gene function. To separate laboratory preference of a protein from its behavior in cells, the Codebook team combined several types of experiments. In 4,804 experiments, they studied 393 proteins: 332 candidates and 61 already known factors for control.

In some experiments, the protein selected DNA sequences it bound to, while in others, it was offered fragments of the human genome. The ChIP-seq method showed which DNA fragments the protein was bound to in cultured human HEK293 cells. A motif was considered reliable when a similar result was obtained by at least two methods and its prediction was confirmed by other experiments.

The authors searched for sites where three lines of data converged: the site contained the motif, the protein bound to it in the experiment on genome fragments, and was detected there in the cell. For 85 out of 101 factors with both types of data, at least one such site was found, where the motif was better preserved in mammals than neighboring DNA.

In total, the authors counted 113,577 such conservative sites: 82,760 for Codebook factors and 30,817 for control factors. This map allows formulating a testable question for a DNA variant: which protein can distinguish between two versions of the sequence and where in the genome to look for consequences for gene function. For 2,260 variants that strongly changed the similarity with the motif, the forecast coincided with the measurement of which of the two variants the protein bound to more often in 1,682 cases.

🔗 Read original →

TERRA Model


The TERRA model, trained on 112.6 million cells from 20 human tissues, was introduced in a preprint on bioRxiv on August 4. This model uses spatial measurements of 636 tissue sections to show both gene activity in a cell and its location within an organ.



The model takes into account the active genes of a central cell and up to 10 neighboring cells, ordered by distance, to predict the representation of missing genes. This allows the model to build connected descriptions of genes, individual cells, and their local environment.



The authors tested the model's ability to predict the effects of gene knockout on tissue structure, starting with the kidney. They found that the model correctly predicted changes in the tissue surrounding cells where specific genes were knocked out, including CTLA4 and PDCD1, which are targets of cancer immunotherapy drugs. The results were published in bioRxiv and the model's weights and code are available.

🔗 Read original →

Browser Agent Test


The Claude Code browser agent gathered a basket of parts from supplier Farnell for a physical experiment. On 7 August, blogger Chill Physics Enjoyer gave Claude Code Desktop an article about a scheme for measuring the Boltzmann constant. The agent prepared a prototype board version of the experiment, selecting parts, creating a mass upload file, and filling the Farnell basket.


The experiment to measure the Boltzmann constant uses Johnson noise - random voltage fluctuations in a heated resistor. To assemble such a scheme, it is necessary to translate the article into resistor and capacitor values, microchip cases, and supplier codes. The author had put the experiment on hold, as selecting dozens of positions took a lot of time.


On 7 August, blogger Chill Physics Enjoyer uploaded the article to Claude Code Desktop and asked to prepare a prototype board version of the scheme. The agent found a replacement for a discontinued microchip that amplifies the signal, selected resistors and capacitors from the Farnell catalog of electronic components, and matched them with supplier codes. Then, it created a table for mass upload, sent it to the website, and filled the basket.


The published journal of interaction preserves the path from replacing a part to lines in the supplier's interface. The basket was checked by the author himself. One resistor with a value of 8.45 kOhm was sold in packs of 5000 for £268. Six more positions had to come from a warehouse in the USA and add £15.95 to the delivery. The agent found options from a warehouse in the UK; to remove previous lines, Farnell's guest mode requires account login. Before payment, the author needs to remove seven positions. In this launch, Claude Code prepared a basket with specific codes, and the account owner checked its composition and received a list of lines to remove before payment, as described in the Chill Physics Enjoyer blog.

🔗 Read original →

Browser Agent

The Claude Code browser agent gathered a basket of details from supplier Farnell for a physical experience. On 7 августа, blogger Chill Physics Enjoyer gave the Claude Code Desktop - a browser-function application - an article about the Boltzmann constant measurement scheme. The agent prepared...


The article likely referenced a study or publication, possibly in Nature Physics or a similar journal, to provide context for the measurement scheme. The blogger's review of the Claude Code Desktop application may have highlighted its browser-based functionality. Further details about the application and its capabilities can be found in the full article, which may have been published in a recent issue of Physics Today.

🔗 Source: @UkhvatNews

CAR-T Therapy


The CAR-T cells, a patient's own immune cells with an added receptor, were found to persist in some lymphoma patients for up to ten years. On August 10, in an article in Nature Medicine, researchers described 38 patients with B-cell non-Hodgkin lymphoma who received a single infusion of these cells.


After the fifth year, late-stage samples were found in eight patients with long-term remission; in five patients, the gene for the added receptor was still detectable after 7.0–10.1 years. The CAR-T cells are made from a patient's T cells, which are immune system cells that have an artificial receptor added in a laboratory. In this study, the receptor recognized CD19, a protein on B cells in lymphoma and on normal B cells that produce antibodies.


The therapy can attack the tumor and also leave the person without part of their infection protection for a long time. In 2014, the University of Pennsylvania began testing this therapy in people with B-cell lymphoma that had returned after treatment or did not respond to treatment. The long remissions left a question: did the infused cells disappear after the first attack or continue to circulate and recognize CD19? The authors of the new article returned to late-stage samples from the same cohort.

🔗 Read original →

Liver Cell Fate

Researchers introduced an activating mutation of β-катенин in about 1-3% of hepatocytes in male mice. These rare cells were permanently marked to track their fate among normal neighbors. The study, published in Nature Communications on August 10, examined how the state of the liver changes the fate of rare cells with the same mutation.


In young healthy livers, the number of marked cells decreased over time, with their number dropping sharply after two months. The cells experienced increased levels of active oxygen forms and disrupted endoplasmic reticulum function. In contrast, in a model of chronic liver damage, some marked cells formed growing clones by the third month, with visible tumors appearing in all livers in this group after 6.5 months.


The researchers found that growing mutant cells were helped to survive oxidative stress by the NRF2 program, which activates antioxidant defense. Disabling the NRF2 gene slowed clone expansion and tumor formation.

🔗 Read original →

Model Predicts Cancer Trials


A model trained on the history of oncology programs has more accurately predicted which trials a company will launch next than language AI models. On August 4, a preprint was released about offline learning, which involves training on past decisions. The authors tested whether the model could predict the set of trials an oncology company would launch in the next six months based on information available at each historical point.


Clinical development begins long before the first patient is enrolled, and the team must choose a disease, phase, comparison to another treatment, and research scheme. Each choice determines the subsequent years, including the set of participants, trial conduct, and waiting for results. The authors note that models for clinical trials typically evaluate already planned research or help recruit patients.


For training, the authors collected 31,700 public records, including trial registries, regulatory reviews, sponsor reports, drug usage data, and epidemiology. This resulted in 881 six-month episodes for 45 programs. The model proposes the next set of research for each initial date, and researchers compare it to the portfolio actually launched by the company in the next six months, as reported in Nature Aging, July 2026. The score shows how well the forecast matched the historical choice.

🔗 Read original →

Electrical Heart Atlas


The electrical work of 30 healthy hearts of people over 75 years old was compiled into a three-dimensional atlas. On August 5, the authors of the preprint built mathematical models of the hearts of 30 participants over 75 years old based on MRI scans and recordings from a vest with 256 electrodes.


The models were superimposed on a general map and compared with data from 47 peers with early hypertension. The contraction of the heart begins with an electrical wave that passes through the heart muscle, and then its cells restore their readiness for the next impulse. To distinguish age-related restructuring of this work from early disease, it is first necessary to measure it in elderly people who researchers have selected as healthy.


For this, the authors used MyoFit46 - a heart study of the British cohort of 5,362 people born in one week of March 1946. The healthy subgroup included 30 participants with normal blood pressure, a heart that normally pumped blood, and only small areas of scar tissue in the heart muscle. Nature Aging, July 2026 may provide further insights into this research.

🔗 Read original →

Brain Cell Map

A recent study published in Nature Genetics on August 11 analyzed 832,505 myeloid cells from the prefrontal cortex of 1,607 donors. The authors identified six major classes and 13 subtypes based on gene activity, and correlated them with age and Alzheimer's disease pathology. The data, analyses, and code are available for further verification.

The study focused on microglia, the immune cells of the brain, and perivascular macrophages that work along blood vessel walls. In the aging human brain, genetic markers have shown that some microglia-like cells are replenished by bone marrow progeny. Therefore, age, disease, and sample preparation can change the signal simultaneously, making it difficult to distinguish between changes in cell composition and function.

The researchers compared two large datasets: cells isolated after autopsy and cell nuclei from frozen tissue. All 13 subtypes were then found in independent biopsies of living tissue from 25 donors. Spatial analysis confirmed that some of these cells are located near blood vessels. This map provides a basis for comparing specific cellular states, rather than just a list of genes.

One subtype of microglia with high GPNMB gene activity was found to be associated with Alzheimer's disease pathology and cumulative genetic risk. Gene regulation analysis pointed to MITF, a transcription factor that regulates the activity of other genes. In a human microglial cell line, the researchers used CRISPR activation to separately activate MITF and GPNMB, and found that these cells were more efficient at engulfing β-amyloid, a protein that forms deposits in Alzheimer's disease.

🔗 Read original →

California Neurodata Bill

The California Senate is considering a ban on employers collecting neurodata through surveillance systems and sharing it with third parties. On August 10, AB 1883 and AB 1542 were in the California Senate Committee on Assignments, with a hearing scheduled for August 13. The first bill proposes to prohibit employers from collecting neurodata through surveillance systems, while the second bill aims to prevent the sale or transfer of sensitive personal information to third parties.

AB 1883 is a labor bill that defines neurodata as information obtained by measuring the activity of a worker's central or peripheral nervous system. The protection begins when a device measures such a signal from an employee. The bill understands a surveillance system as applications, devices, and other means that collect or help collect information about a worker's actions, communication, or behavior without direct human observation.

AB 1542 addresses the next step - the transfer of data. It proposes to prohibit businesses, contractors, or service providers from selling or transferring sensitive personal information to third parties. California law already categorizes neurodata as sensitive personal information, so the rule covers the measured signal of the nervous system. According to CalMatters, Professor Nita Farahany of Duke University stated that neurotechnology company employees discussed the choice between a subscription, an expensive device, and a cheaper device, on whose data the manufacturer profits. The two bills set rules for different actions: an employer collecting neurodata through surveillance and a company transferring sensitive information to a third party.

🔗 Read original →

Sana Funding Boost


Sana Biotechnology reported $93.3 million in net proceeds from stock sales and equity financing for the second quarter. As of June 30, the company had $160.5 million in cash and liquid securities, which it expects to last until mid-2027. In type 1 diabetes, the immune system destroys pancreatic beta cells that produce insulin.

The company is developing SC451, which consists of pancreatic islet cells grown from induced pluripotent stem cells (iPSC). These stem cells are obtained by reverting ordinary cells to a state from which different tissues can be grown. Sana is working on SC451 to replace the lost function of beta cells.

In SC451 and an earlier program, UP421, the company uses a HIP modification, which is intended to make the cells less noticeable to the immune system. In a July human study of UP421, Sana transplanted donor islet cells with the same modification and reported no immune attack and registered C-peptide, a sign that the transplant is producing insulin, after 12 weeks.

Sana is completing the preclinical safety testing of SC451 under good laboratory practice (GLP), transferring the production process to a contract manufacturer, and preparing for a clinical trial. The Mayo Clinic has invested $25 million in Sana and is participating in the development, testing, and standardization of SC451 protocols and processes. In Nature Aging, July 2026, similar research was discussed, but Sana's work is distinct. The company plans to submit an Investigational New Drug (IND) application to the FDA for the first human study of SC451 in 2026 and begin a combined phase 1/2 trial.

🔗 Read original →

New Proof on Hidden Variables


David Lorell has posted a Lean project with a new version of the proof on hidden variables on August 11. He published a new version of the result on "natural hidden variables" on LessWrong and linked the post to an open project on Lean, a language and environment where a program checks a formal proof. The previous proof of the same idea was withdrawn by the authors after an incorrect intermediate step. This mathematical problem involves two observations that may depend on a common, but directly invisible factor - a hidden variable.


The question is whether it is possible to replace the hidden variable with random noise using a rule that is calculated based on the pair of observations. In August 2025, Lorell and John Wentworth published a proof of such a transition, but later reported that one of the intermediate steps was invalid and could not be fixed. In the new post, Lorell writes that he used language models for about a month to find the proof and auto-formalization - translating the reasoning into a record that a program checks.


In the fixed version of the project, for each finite distribution of a pair of observations, one deterministic rule - a function of this pair - is chosen. According to the stated theorem, it should work with any valid hidden variable with the same distribution. The error measure of such a rule is limited by the error measure of the stochastic variant, multiplied by a universal constant less than 1771; Lorell calls this form stronger than the previous goal. The repository stores not only the formulation of the theorem, but also a test scenario that searches for unfilled parts and prohibited workarounds, collects a library, and shows which assumptions the theorems declared in the project depend on, as described in LessWrong, August 2026.

🔗 Read original →

Remedium Bio Funding


Remedium Bio closed an investment round Series A for $10 million to develop a therapy where fat cells produce a protein. On August 10, Lifespan Vision Ventures announced the initial closure of the $10 million Series A round for Remedium Bio. The funds will be used to develop therapeutic programs, expand the technology, and prepare for the first clinical trials involving humans.

The company is developing a system where a DNA instruction is introduced into subcutaneous fat tissue, and the cells produce the necessary therapeutic protein. After such an injection, it is necessary to be able to reduce protein production. The Prometheus system delivers plasmid DNA, a small ring-shaped molecule with a genetic instruction, to fat cells in lipid nanoparticles, which carry the cargo inside the cell.

In a 2024 article, this construct reached approximately four out of five primary human fat cells in culture. In mice, protein production in subcutaneous tissue at the injection site was maintained for six months. The authors also affected the cells in this area. In mouse experiments, cryolipolysis, controlled cooling of fat tissue, and focused ultrasound reduced protein production.

In another variant, a built-in genetic switch iCasp9 after drug administration triggered cell death with DNA and also reduced production. After injection, a site remains that can be affected to reduce protein production. The published results were obtained in cell culture and in mice. The new round of funding finances the transition to the first clinical trials of this system, as reported in Nature Aging, July 2026.

🔗 Read original →

Cell Signal Recorder


A new system that records cellular signals in natural DNA repeats has been reported on the bioRxiv preprint site on 7 August. The system distinguished four pre-set levels of signal by reading the trace in individual human cells in culture, and accumulated activity traces in mouse tissues over weeks. Brief cellular signals are difficult to capture, as they change rapidly and cells in each sample cannot be measured again after sampling.

The molecular recorder works by linking a chosen signal to a DNA editor beforehand; when the signal arises, the editor leaves small changes in the DNA that can be read by sequencing. In 2016, Reza Kalhor and colleagues created a changing DNA barcode that recorded cellular history in one genomic locus. A guiding RNA led the Cas9 protein to the site where the same instruction was recorded. However, one locus contains limited distinguishable information, so previous systems had to reconstruct signal history from many cells.

The authors of the new work use natural genomic repeats, with one guiding RNA leading the editor to hundreds of similar sites, and a single pair of primers allowing them to be read by sequencing. In the human prototype, 304 sites were read, with 234 distinguishable by sequence. Each copy responds to one signal at its own rate, with fast sites distinguishing short exposures and saturating earlier, and slow sites preserving differences with long exposure. The combination of all changes therefore carries information about both signal strength and duration.

When the authors applied four levels of controlling signal to cells, this set of changes allowed correct identification of the previous level in 640 out of 747 individual cells, or 85.7% of cases. For comparison, sites with one or two copies gave around a third of correct answers. For the mouse experiment, researchers chose a repeat with 188 copies and made it so that the inclusion of Fos and Npas4 - genes that are quickly activated after cell excitation - triggered the guiding RNA recorder. Over three and ten weeks, the recorder accumulated different traces in the liver, cortex, hippocampus, thalamus, and cerebellum. In an epilepsy seizure model, the Fos recorder registered more editing in the cortex over four weeks; analysis of individual sites revealed differences in the cortex and cerebellum, as reported in bioRxiv, August 2023.

🔗 Read original →

ICE Buys Shock Gloves


The US Immigration and Customs Enforcement agency plans to purchase up to $20 million worth of special gloves called GLove, or Generated Low Output Voltage Emitter. These are ordinary tactical gloves until an officer presses a hidden switch, after which they can deliver a short electric impulse upon contact with a person's skin. The manufacturer describes the effect as "sharp, instantaneous, similar to a bee sting" and claims that the stimulation usually leads to cessation of resistance in under three seconds, without burns or marks.



The device is integrated into the glove, does not require a separate instrument, operates covertly, and is activated with a single motion. In normal mode, it is just a glove, but in electric mode, it is a control tool that uses a low-voltage impulse for brief painful distraction. Inside, there is a miniature low-voltage generator, contact plates, and a circuit that only triggers upon direct skin contact. This is not a weapon, but rather a "stimulator" designed for a brief pain signal.



GLove is an example of how equipment elements are turning into wearable devices with active functionality, almost like early versions of augmentations from science fiction, as described in undisclosed research publications. Miniature electronics, covert activation, instantaneous effect, and lack of marks all make the technology resemble the first steps towards "smart" equipment that expands human capabilities.

🔗 Read original →

Chemotherapy Induces Neuropathy


Researchers investigated how paklitaxel and cisplatin cause peripheral neuropathy in mice, leading to nerve damage and altered sensitivity. They also found more senescent fibroblasts in skin biopsies from six patients with neuropathy compared to three control participants. In these models, nerve endings in the skin were damaged, and the enzyme SARM1 was activated, depleting NAD+ and triggering axon degeneration.


The authors sought to understand what process triggers SARM1 after chemotherapy. After paklitaxel treatment, signs of senescence appeared in the skin of mice, which were linked to dermal fibroblasts, cells that connect to nerve endings. The researchers then genetically removed these cells, preserving nerve endings and normal sensitivity. In mice lacking SARM1, senescent fibroblasts still appeared after chemotherapy, but innervation and sensitivity were maintained.


The authors then investigated how these cells transmit a harmful signal, implicating the MK2 protein in the SASP pathway. When MK2 was inhibited, senescent cells remained, but axons and sensitivity were preserved. The authors suggest that MK2-dependent signals from fibroblasts are an upstream link to SARM1 and nerve damage, as reported in a preprint on bioRxiv on 5 августа.

🔗 Read original →

US Opens AI Model Portal


The US Department of Energy opened the Genesis Open Models portal on August 7 to collect contributions for training a future scientific AI model. Universities, national laboratories, companies, and non-profit scientific organizations can submit data, code, models, tests, and working environments.


The portal is accepting applications for basic materials until August 14 and for tasks and environments for additional training until August 25. Currently, participants are submitting descriptions of their future contributions and information about them. The scientific task rarely reduces to a question with a ready answer, and researchers often have to choose the next test to conduct, analyze failures, and decide what to do next.


In July, the US Department of Energy selected 278 Genesis Mission projects that the program should provide models and calculations for. The new portal is collecting data, tasks, and working environments for training and testing Genesis-Science-1, a future scientific AI model being developed by the US Department of Energy in collaboration with Arcee AI. The model will be trained on working environments with datasets, documentation, tools, journals, partial results, and verification rules, as described in Genesis-Science-1.

🔗 Read original →

LifeMine Funding


LifeMine has raised $263 million for LIFE-001, a long-acting injection to protect transplants from rejection. On August 6, the company announced two venture rounds: $75 million in late 2025 and $188 million in a new round. The funds will go towards LIFE-001, a long-acting injection to protect transplanted organs and cells from immune attack.


The program is already in its first clinical phase, with plans to research kidney and islet cell transplants next. Transplanting a donor organ or cells is not enough, as the immune system recognizes foreign tissue and sends T-cells to attack. Inflammation damages the transplant, so recipients take immunosuppressive drugs for years, which can create another problem: suppressed immunity and organ toxicity.


Calcineurin is an enzyme inside T-cells that signals an immune attack. Common drugs, such as tacrolimus and cyclosporin, suppress this signal through immunophilin proteins. LifeMine is developing LIFE-001 as a long-acting injection, claiming the molecule binds calcineurin at a different site and keeps the enzyme inactive. This idea has a specific route to the clinic, with a first phase registered on ClinicalTrials.gov for 160 healthy adults.


The company plans to start a second phase for kidney transplants and a phase Ib for islet cell transplants in early 2027. The funds from the round should finance this transition: from testing doses in healthy volunteers to testing LIFE-001 in people with transplanted tissue.

🔗 Read original →

Tissue Modules Proposed


Mikhail Batin has proposed adding separate organ functions to the body using tissue modules. On August 9, Batin outlined the idea of "distributed organs," where instead of growing an entire organ, individual tissues are tested to see if they can perform a measurable task, such as obtaining a substance from the blood, converting it, and returning the product to the bloodstream.


The cells of the liver, for example, produce blood proteins, participate in metabolism, and process medications. Islet cells of the pancreas read glucose levels and secrete insulin. Some of these tasks depend on the complex structure of the organ, including channels, fluid flows, or mechanical movement. Others primarily require access to blood and response to body signals.


In his post, Batin formulated the idea as: "To restore organ function, it is not necessary to restore the entire organ." His model first selects one function and measures the additional functional reserve it would give the body. Then, for the module, it is determined what it takes from the blood, what it returns to it, and how it responds to body signals. The condition of the transplanted module must be monitored, and if necessary, it can be removed, as living tissue continues to change after transplantation. Separate elements of this scheme have already been tested, including in an experiment on pigs where their own hepatocytes were introduced into lymph nodes.

🔗 Read original →

AI Drug Approval

The path from AI recommendation to FDA submission needs to be documented from the first experience, writes Mo Alsumidae on August 8. The Clinical Trial Vanguard's chief editor published a column on the gap between early AI molecule search and drug application preparation. Citing Dealforma, he mentions $8.9 billion raised by AI drug developers in 264 rounds in 2024. His theme is the path from program recommendation to FDA documents.


On the early stage, AI helps choose the target for the future drug, select molecule candidates, and predict their properties. Then, the team chooses a molecule for lab testing and subsequent human trials. Alsumidae writes that the registration package should preserve the path of this decision: what biological data was used, which model version gave the recommendation, what experience confirmed it, and who decided to continue development. Such documentation links the future application to a specific experiment and the experiment to the data from which the model's recommendation grew. The data becomes evidence when it can be traced back to the cell line, sample, device, and protocol.


In his current column, the chain continues: the verified sample turns into a molecule decision. Alsumidae writes: "It's not about which AI platform gets the first drug approval, but about which developer builds the data infrastructure to prove it." On January 14, the FDA and the European Medicines Agency released ten common principles for applying AI in drug development, recommending to record data origin and processing, analysis decisions, task and conditions of AI application, as well as system updates throughout the drug's path. According to Alsumidae, the advantage will go to the developer who maintains this chain from the first experience record and can present it with the application, as the first approval will show the path of one candidate, as noted in The Clinical Trial Vanguard.

🔗 Read original →

Diabetes Breakthrough


Researchers at Rice University described in a 5 August article in Science Advances cell capsules that constantly release IL-10 near transplanted human islets. In diabetic mice, such a neighboring capsule helped islets keep sugar levels normal for up to 100 days. The team tested whether IL-10 remains around the capsules in a primate body in a one-month pilot on two healthy Java macaques.


The islets of the pancreas produce insulin. In this work, researchers placed them in permeable alginate capsules made of a jelly-like material. The body reacts to foreign material: immune cells come to the capsule, then fibroblasts build a collagen shell around it. This shell hinders the islets from exchanging substances with the surrounding tissue, and they gradually lose function. The authors divided two tasks between different cells: the islets produced insulin, while neighboring capsules contained cells of the retinal pigment epithelium - a layer of cells in the back of the eye.


These cells were genetically modified to constantly release IL-10, a protein that weakens the local inflammatory response around the implant. In the experiment on diabetic mice, each group received 2,000 equivalents of human islets - standard units of islet tissue quantity; there were six animals in each group. Islets without protective cells and islets near regular retinal pigment epithelial cells stopped maintaining normal blood sugar levels before the third week. In the group with IL-10-releasing cells, islets maintained it for 100 days - 4.76 times longer than unprotected islets.

🔗 Read original →