A dependency screen nobody has to run, and a resistance model that inverts across town

A transformer pretrained on gene-essentiality scores now reads an ordinary transcriptome and returns an inferred vulnerability profile, so the expensive screen is needed for training and not at the moment of use. A review of virtual cells published the same day reports that independent benchmarks find scale alone does not buy generalisation outside the training distribution. A model that predicts antibiotic resistance from routine hospital mass spectra does worse than fail at a second hospital: it attaches resistant isolates to the wrong neighbours and carries the wrong prior with them. India capped distributor margins on 137 anti-cancer drugs, patented ones included, with new ceiling prices from 9 November. Anthropic committed a twentieth of its shares to a public benefit fund held in a wholly owned company rather than a charity.

Connecting the dots

A cheap measurement standing in for an expensive one, and the test that would settle it. The vulnerability transformer's whole move is that "CRISPR dependency scores were used only as pretraining targets and were not required as model inputs", and its authors say plainly that the output "represents a CRISPR-supervised transformation of baseline expression rather than a direct CRISPR measurement". The virtual-cell review, "Virtual cells: progress and challenges from mechanistic modeling to data-driven integration", says of this whole class of models that "parameter count and training-set size do not automatically yield robust out-of-distribution generalization". Read together, the interesting question about an inferred dependency profile is which split it was tested on, and both papers come with that question attached.

Two literatures solving one transfer problem. The resistance paper finds that "site-level covariate shift inverts neighborhood attachment", which is a failure of a model crossing a hospital boundary. In the same day's journal lists sit "Raman spectroscopy coupled with interpretable machine learning for early detection of citrus Huanglongbing" and "Multi‑cultivar NIR determination of soluble solids content in cherry tomatoes", a literature whose daily business is transferring a calibration between instruments and cultivars. The vocabulary differs; the instrument-to-instrument problem is the same one, and it has forty years of chemometrics behind it.

Legibility, in a week when three groups were building registers. James Scott's sentence is that "Certain forms of knowledge and control require a narrowing of vision." In Bamako a founder of a decade-old digital-health observatory says identification "fait partie des fondamentaux qui ne sont toujours pas mis en place" (is one of the fundamentals still not in place), so one malaria episode seen at three facilities is three cases. In Dar es Salaam a developer put the national treatment guideline on a model hub and had to add version control, because the ministry re-issued the source document. The register is being built at both ends, and the first thing it needs is a way to say which version of the world it depicts.

Agents at the bench, and what counts as a result. A benchmark of fifteen open scientific problems reports that "None of the 135 submissions satisfies all the conditions, and 114 fail our integrity check," with failures mostly of work claimed and not done. A lab-automation project released the same weekend takes the opposite approach and makes the unit of progress the permission: reads run free, "anything that moves or reconfigures stops for a person."

New concepts

A brain's shape spectrum, truncated, is a clock

"the maximum harmonic degree acts as a developmental coordinate" — Goldschmidt, bioRxiv

"the truncation degree of the spherical-harmonic expansion is, to a first approximation, a clock" — same

The fetal cortical surface is treated as a band-limited object, expanded in spherical harmonics and reduced to a short vector of band powers, from which age falls out by nearest-neighbour regression with no learned model anywhere in the pipeline. The claim with consequences is that a low-degree version of a mature cortex is a younger brain, which folds folding and spectral bandwidth into one axis. Off-trajectory distance rather than size is what the author proposes as the abnormality signal.

Why it matters: a developmental timebase computed from shape alone, with no molecular measurement and no training set, for anyone who needs to place an individual brain on a trajectory.

The spinal cord gets a manifold, and the circuit diagram is demoted

"For decades, spinal motor function has been interpreted either through the lens of alternating flexor/extensor modules" "or through genetic characterization of interneurons." — Berg et al., Current Opinion in Neurobiology

"Both approaches struggle to amalgamate and explain how movement is flexibly generated, adapted, and stopped" "without losing posture." — same

"the spinal cord, which houses the core circuitry responsible for rhythmic motor output such as walking," "has remained largely unexplored from a manifold perspective." — same

The review argues the population-dynamics framework that reorganised cortical motor control is "a complementary paradigm" for the cord, where the traditional module-and-interneuron account struggles with starting, adapting and stopping movement while posture holds.

Why it matters: the vocabulary of state space arrives in a system where the wiring is unusually well characterised, which makes it a fair place to ask what the manifold adds to the circuit.

Basic biology

One of three related cytokines does nearly all the work in human pain neurons

"LIF alters the transcriptome of sensory neurons more prominently than IL-6 and IL-11" — Ross et al., bioRxiv, CC-BY 4.0

"LIF is the most consequential cytokine out of the three" — same

"intraarticular injection of LIF into a healthy mouse knee resulted in" "increased responsiveness to painful pressure stimuli in that joint" — same

Leukaemia inhibitory factor, interleukin-6 and interleukin-11 all signal through the same receptor family and the same downstream kinases, and in human stem-cell-derived sensory neurons they are not interchangeable. The ligand is "consistently present in the SF from painful joints from individuals with inflammatory arthritis", where SF is synovial fluid, and injecting it into a healthy joint is enough to produce pressure sensitivity.

Why it matters: for rheumatology, a candidate explanation of why pain persists when inflammation is controlled, and a specific ligand to block rather than a receptor shared by three.

Which duplicate opsin a cone expresses is set by hormones, and can change later

"A foundational model posited that regulation of human LWS vs. MWS opsin expression is stochastic," "whereby an upstream enhancer associates randomly with promoters of the proximal or distal gene," "and one of these associations becomes permanent." — Stenkamp, Progress in Retinal and Eye Research

"differential expression of these genes is regulated by the paracrine signal retinoic acid (RA)" "and the endocrine signal thyroid hormone (TH)." — same

"plasticity of cone opsin expression across the lifespan in response to TH" — same

The primary result the review is built on is a decade old and open: "This is the first evidence that an extracellular signal may regulate differential expression of opsin genes" in a tandemly duplicated array, with "individual cone photoreceptors were induced by RA to switch expression from LWS2 to LWS1" — Mitchell et al., PLoS Genetics 2015, CC BY.

Why it matters: a worked case of how duplicated genes acquire divergent expression, with external signalling onto shared regulatory DNA in place of a coin flip, and an identity that stays revisable in a cell that will never divide again.

Atlases and datasets

Nucleosome positions across four hundred tumours, read out of chromatin-accessibility data

"We systematically characterized nucleosome organization across 410 primary tumors spanning 23 cancer types using ATAC-seq" — bioRxiv, CC-BY-NC-ND 4.0

"We also developed OncoNucAtlas, a publicly accessible database facilitating exploration of these nucleosome landscapes." — same

The pan-cancer resource is derived from assays that were run for a different purpose, and the stated application is cell-free DNA detection, where nucleosomes sit determines where circulating fragments break. The abstract announces the database without an address for it.

Why it matters: a reference map of nucleosome organisation by tumour type is the interpretation layer that fragment-based blood tests have been missing, once the database can be reached.

Methods

Functional vulnerability becomes something you infer from expression alone

"Here, we present vulnerability-guided cascaded attention transformer (VCAT), which integrates" drug-perturbation signatures with CRISPR-supervised cellular representations — Chen et al., Briefings in Bioinformatics, CC BY

"CRISPR dependency scores were used only as pretraining targets and were not required as model inputs" — same

"Ablation analyses showed that CRISPR-derived vulnerability features complement baseline gene expression" — same

The model is pretrained to predict gene-effect scores from baseline expression, so at inference a sample with only a transcriptome carries a dependency profile and no screen is run. The authors state the epistemic status of that profile themselves: it "represents a CRISPR-supervised transformation of baseline expression rather than a direct CRISPR measurement". Code and checkpoints are released, and three genes picked out by the attention layer were knocked down at the bench to test whether they change sensitivity to a liver-cancer drug.

Why it matters: for anyone who cannot run a genome-scale screen on their own samples, the screen becomes a training asset owned by somebody else and the measurement becomes bulk expression.

Figure 1, Chen et al., Briefings in Bioinformatics 2026, CC BY, https://doi.org/10.1093/bib/bbag559

An off-the-shelf cell therapy that leaves the T-cell receptor in place

"Here, we developed a next-generation UCAR-T strategy that avoids TCR/MHC disruption" — Li et al., Molecular Therapy

"First, we eliminated CD4/CD8 to reduce alloreactivity while preserving TCR integrity" — same

"we identified CD70 as a marker selectively upregulated on alloreactive immune cells" — same

Both standard edits are inverted. Instead of deleting the T-cell receptor to prevent graft-versus-host reactivity the group deletes the CD4 and CD8 co-receptors, and instead of hiding from the recipient's immune system by removing the major histocompatibility complex it points a receptor at the rejecting cells. The product then has to be protected from itself, so "we additionally deleted CD70 to prevent fratricide". The anti-rejection half of the design is prior art: another group reports "an optimized CD70 CAR that prevents rejection of allogeneic CAR T cells by targeting activated alloreactive lymphocytes" — Zhang et al., Nature Communications.

Why it matters: donor-derived cell products lose persistence when the receptor is cut out, and this is a design that keeps it; protection from graft-versus-host disease in an animal is the claim still owed.

An antibody carries small interfering RNA past the blood-brain barrier

the conjugate "demonstrated enhanced penetration into central nervous system (CNS) tissues compared to unconjugated siRNA" — Tian et al., bioRxiv, CC-BY-NC-ND 4.0

it "achieved widespread knockdown in muscle, heart, and lung." — same

The vehicle is an antibody against the insulin-like growth factor 1 receptor, given intravenously, aiming at the problem that brain-directed oligonucleotide drugs still mostly go in through the spine.

Why it matters: another receptor added to a very short list of routes into the brain for nucleic-acid drugs, with lung knockdown after systemic dosing as the unexpected part.

Clinical trials

Two Japanese encephalitis vaccines can be mixed mid-schedule

"Mixed-product JENVAC and JEEV schedules were non-inferior to same-product schedules" — EClinicalMedicine

Nine hundred infants in India were randomised across four two-dose schedules, with seroconversion by neutralisation test after the second dose close to ninety per cent in every arm. The trial was funded by the country's own medical research council.

Why it matters: a programme tested whether its two locally used products are interchangeable when supply fails, which removes a logistical constraint rather than adding a product.

Home cardiac rehabilitation matched the hospital programme at eight weeks

At a Spanish public hospital, 129 patients after acute coronary syndrome were randomised to telerehabilitation or centre-based rehabilitation, and 86 per cent against 79.7 per cent met a composite of seven cardiovascular and lifestyle goals, a difference with a p-value of 0.37 — Eur Heart J Open.

Why it matters: a facility-based programme replaced by a remote one inside a public system, with the short-term risk-factor composite unchanged.

Adding octreotide to acid suppression in ICU gastrointestinal bleeding

In a small randomised trial in Egypt, rebleeding, transfusion and ventilation were no different, while in-hospital mortality was 39.3 per cent against 7 per cent in favour of the proton pump inhibitor alone, p=0.011 — Am J Med Sci.

Why it matters: fifty-six patients and an unadjusted mortality difference is not a verdict on the drug, and it is the kind of signal that decides whether a larger trial happens.

Hospitals

Korea: intensive care below the top tier has not moved in four years

(translated from Korean) "Of 258 general hospitals in the 4th ICU adequacy evaluation in 2023, 97, or just 37.6%, operated dedicated intensivists." Original: "2023년 제4차 중환자실 적정성 평가 대상 종합병원 258곳 중 전담전문의를 운영한 기관은 97곳으로 37.6%에 불과했다." — Korean Doctors' Weekly

(translated from Korean) "The general hospital rate for dedicated intensivists was also 37.6% in the 3rd evaluation in 2019, showing no improvement across the two evaluations." Original: "종합병원의 전담전문의 운영률은 2019년 제3차 평가에서도 37.6%로 나타나 두 차례 평가에서 개선되지 않았다." — same

Nurses in general-hospital intensive care units carry more than twice the beds their tertiary-hospital colleagues do.

Why it matters: these are the hospitals that tertiary centres refer patients back to, so the referral chain ends in the tier whose critical care has not improved between two national assessments.

Korea: nine in ten health-sector intrusions hit the smallest providers

(translated from Korean) "Of 28 breaches reported at private medical institutions from January to September 2026, 25, or 89.3%, occurred at hospitals and clinics." Original: "2026년 1~9월 민간 의료기관에서 접수된 침해사고 28건 중 25건에 해당하는 89.3%가 병원과 의원에서 발생했다." — Korean Doctors' Weekly

(translated from Korean) "The main intrusion routes were ports opened externally for remote maintenance and vulnerabilities in remote-control programs." Original: "주요 침해 경로는 원격 유지보수를 위해 외부에 개방한 포트와 원격제어 프로그램의 취약점 등이었다." — same

Of more than seventy-five thousand clinic-level facilities, five are enrolled in the national monitoring service.

Why it matters: the door is the vendor's maintenance connection, which means the exposure belongs to whoever services the software rather than to the clinic that bought it.

England: care in corridors, without a definition or a clock

"in March 2025, 10,042 (17.7%) patients received care in a temporary care environment" — HSSIB

"National and local data on the time patients are in a temporary care environments is variable and inconsistent." — same

"The definition of temporary care environments varies across organisations" — same

The investigators found risks including monitoring difficulty, staffing ratios that could not be met, and "A lack of piped oxygen and suction."

Why it matters: the national safety body's main ask is for agreed definitions and data collection, because the practice cannot currently be counted, let alone compared between hospitals.

Medicine concept of the day

The Glasgow-Blatchford score, and what a rule-out is for

The score is computed before anyone looks inside the gut. Its inputs are bedside and laboratory items available at the door: "We calculated risk scores from patients' admission haemoglobin, blood urea, pulse, and systolic blood pressure" "as well as presentation with syncope or melaena, and evidence of hepatic disease or cardiac failure", and what it predicts is an action rather than a diagnosis, "a risk score that predicts patients' risks of needing blood transfusion or intervention", with a ROC curve area of 0.92 in the derivation paper — Blatchford et al., Lancet 2000. Urea is in the score because digested blood is absorbed as protein, and transfusion sits in the outcome while haemoglobin sits in the inputs.

The comparison across three thousand consecutive patients at six hospitals in four regions settled which score to use and what it cannot do:

"The Glasgow Blatchford score was best (area under the receiver operating characteristic curve (AUROC) 0.86)" "at predicting intervention or death" — Stanley et al., BMJ 2017

"A Glasgow Blatchford score of ≤1 was the optimum threshold" "to predict survival without intervention" — same

"No score was helpful at predicting rebleeding or length of stay." — same

Why it matters: it is a discharge instrument, and the two outcomes it does not predict, rebleeding and length of stay, are exactly the outcomes today's intensive-care bleeding trial measured. A score that identifies who can go home only pays off where going home is an option and a next-day endoscopy slot exists.

Worth knowing

Mast cells confirmed as drivers of asthma remodelling, with the tool compound withdrawn

blocking the mast-cell survival receptor "prevented HDM-induced bronchoconstriction and airway hyperresponsiveness" and "largely attenuated HDM-induced increases in mast cells, inflammation," "airway smooth muscle layer thickening, and collagen deposition" — Xiang et al., J Allergy Clin Immunol

"Mast cell-dependent signaling is a key driver of allergic asthma pathogenesis." — same

The molecule that proved it is a selective KIT inhibitor whose trial in chronic cold urticaria is listed as terminated, after "moderate drug induced liver injury in two subjects enrolled in the first dose cohort" — NCT05510843.

Why it matters: target validation and a usable drug are separate achievements, and the registry holds the half the paper does not report.

A map of where virtual cells actually are

the review uses "state encoding (E), state transition (T), and observation decoding (D) as a conceptual abstraction" across mechanistic models, foundation models and perturbation models — Ma et al., Briefings in Bioinformatics, CC BY

"E/T/D is therefore a conceptual comparative abstraction rather than a mandatory architectural decomposition." — same

"parameter count and training-set size do not automatically yield robust out-of-distribution generalization" — same

Why it matters: a comparative frame that puts differential equations and learned models on the same axes, with a stated limit on what scaling has bought so far.

Figure 2, Ma et al., Briefings in Bioinformatics 2026, CC BY, https://doi.org/10.1093/bib/bbag562

Health systems and policy

India caps the distributor's cut on cancer drugs

"The government has capped trade margins at 30% for 137 non-scheduled anti-cancer drugs" — Kallianpur, The Oncology Margin Cap

"The revised maximum retail prices (MRPs) take effect on 9 November 2026." — same

The order reaches patented products that scheduled price control does not, and the author's own caution is that "MRP is not necessarily the price at which every medicine is sold."

Why it matters: the intervention is on the distribution markup rather than the manufacturer's price, which is a different lever from compulsory licensing or scheduled ceilings.

Ten years of digital health in West Africa, without a patient identifier

(translated from French) "the advances proposed are sometimes disconnected from the real needs of patients and health professionals." Original: « les avancées proposées sont parfois déconnectées des besoins réels des patients et des professionnels de santé. » — RFI via allAfrica

(translated from French) "Identification is one of the fundamentals that are still not in place." … "our data are of poor quality." Original: « L'identification fait partie des fondamentaux qui ne sont toujours pas mis en place. » … « nos données sont de mauvaise qualité. » — same

The example given is one patient seen at three facilities for the same malaria episode and counted three times. The success named is a teledermatology network covering the interior of Mali from more than a hundred sites.

Why it matters: without an identifier, surveillance counts visits and not people, which is upstream of every incidence estimate built on those records.

Onset ages for mental health conditions, condition by condition

"Around a third of people who will ever experience a mental health disorder do so by the age of 14." — Ritchie, Our World in Data, CC BY

for social anxiety, "almost 80% of those who will experience it at some point in their lives do so before the age of 18." — same

"Knowing the age at which people might be particularly vulnerable to the onset of a mental health condition is important." — same

The underlying meta-analysis pools nearly two hundred epidemiological studies, mostly from high-income countries, and the peaks differ sharply by condition.

Why it matters: prevention programmes have to pick an age, and this gives a different age for eating disorders than for schizophrenia.

Ebola: Kenya's eighth case is support staff, and a ninth province reports in the DRC

"who worked as a patient attendant cleaning utensils and linens" and "was likely exposed to Kenya's first Ebola case, discovered earlier this month." — Al Jazeera

"The National Institute for Biomedical Research confirmed two cases – including one death – in North Ubangi." — TRT Afrika

The DRC situation report cited by the same outlet brings the cumulative total to "9,415 cases and 4,536 deaths since the outbreak began."

Why it matters: in Kenya every case so far was already in quarantine, and the exposed group now includes people who handle linen rather than patients.

Patents

Eligibility is back to a fifth of the appeal docket

"I have been parsing ex parte PTAB decisions - about 130 k for this post running from 2009 through September 2026" — Crouch, Patently-O

"eligibility is again about a fifth of the PTAB's appeal docket" — same

The board has nearly stopped adding eligibility rejections of its own, after "Director Squires, writing for an Appeals Review Panel, vacated the Section 101 rejection." in a case about machine-learning claims.

Why it matters: subject-matter eligibility is the gate for diagnostic-method and software claims in the United States, and examiners are still sending a fifth of appealed cases through it.

Startups and deals

A company whose product is the perturbation data

(translated from Chinese) "Liman Bio focuses on high-throughput cell editing and perturbation-sample data production, and aims to become the upstream data-infrastructure supplier of the AI virtual cell industry." Original: "粒曼生物专注于高通量细胞编辑与扰动样本数据生产，致力于成为人工智能虚拟细胞（AIVC）产业上游的数据基础设施供应商。" — VCBeat

(translated from Chinese) "The Liman team has completed the construction and wet-lab validation of an arrayed perturbation library covering 19,883 human genes." Original: "粒曼团队已完成覆盖人类19,883个基因扰动阵列式文库的建设与湿实验验证。" — same

Why it matters: the second Chinese angel round this month for selling data rather than models to the virtual-cell field, which is the supply side of every paper in that literature.

Midea takes control of Esaote

"Midea Group (美的集团) completed its acquisition of a controlling stake in Italian medical imaging company Esaote S.p.A." — EqualOcean

"Esaote reported fiscal 2025 revenue of €267 million, with about 70% generated in international markets." — same

Why it matters: a Chinese appliance group now controls a European maker of mid-range ultrasound and magnetic resonance systems, the imaging class most likely to be bought by a district hospital.

Repos and tools

Micron positioning from printed flexures, at hobby prices

RepRapMicron is "an extension of the Open Source RepRap 3D printer project" and "aims to bring micron-scale fabrication into very widespread adoption." "Range of motion is approximately 8mm on X, Y, and Z axes." — VikOlliver/RepRapMicron, GPL-3.0

a second project is "an open source low-cost, easy-to-build motorized XYZ Micro-Manipulator motion control platform achieving submicron precision" for "optical alignment, probing electronic components, and microscopy", with "steps down to 50nm" and the warning "Please mind the difference between resolution and accuracy." — 0x23/MicroManipulatorStepper

Why it matters: positioning stages in the micron class, printable and documented, for labs that cannot buy a commercial micromanipulator.

A permission layer between an agent and the instruments

"Turn your whole lab — hardware and software — into safe, typed, permission-gated tools for an AI agent." with the rule that "reads run free; anything that moves or reconfigures stops for a person." — determlab/shal, MIT

The project reports that "In a blind test, an agent wrote working, safety-checked drivers for 4 of 4 devices from documentation alone." and says of itself "SHAL is in alpha (Phase 1)." with the result "Founder-run, not yet independently replicated."

Why it matters: the model-to-instrument plumbing now has a stated safety semantics, where the side effect of an operation decides whether a human approves it.

AI: using it well

A benchmark where agents mostly failed the integrity check, not the science

"Can AI agents carry out research on open scientific problems?" — "Into the Unknown", Carnegie Mellon and national laboratories

"None of the 135 submissions satisfies all the conditions, and 114 fail our integrity check," — same

Fifteen open problems in quantum computing, plasma physics, chemistry, materials and scientific machine learning were each given a rubric written by a working scientist, and most failures were papers claiming work that was not done or reporting numbers the logs contradict.

Why it matters: the result relocates the bottleneck for autonomous research from capability to verification, and it is the argument for designs where every claim is checked against an artifact.

Other frontiers

Six hundred photonic cavities stamped onto diamond

"They used this process to fabricate more than 600 GaP nanophotonic cavities on a diamond substrate with a 99.5% yield." — Phys.org on Yama et al., Optica Quantum

"The devices also had spin-relaxation times exceeding 0.4 milliseconds" — same

Gallium phosphide cavities are stamp-transferred onto diamond holding silicon-vacancy centres, with enough coupling to read a single spin from scattered light in one shot.

Why it matters: the headline number is a manufacturing yield rather than a fidelity, in the same material system that biology uses for diamond magnetometry.

A Wi-Fi dongle as a software-defined radio

"The latest chip to join their ranks is the MediaTek MT7921AU, often used in Wi-Fi adapters" — Hackaday

"BirriaSDR instead streams these to the host at 40 Msps in 14-bit resolution, with the option to tune from 2156 to 2610 MHz." and "it doesn't make any permanent firmware changes, so the Wi-Fi adapter goes back to its original state" — same

A second project reaches "a maximum bandwidth of 80 MHz and tune to the 5 GHz bands."

Why it matters: a commodity part becomes an instrument by reading registers the vendor already exposed, with no hardware change and no new supply chain.

Big tech

Anthropic puts equity into a fund it owns

"Subject to stockholder approval, 5% of Anthropic's shares will be committed today." with "an additional 2% added at the board's discretion for every 50% increase in the company's value above a set baseline." — Anthropic

the fund is established "as a wholly-owned LLC—rather than a 501(c)(3) or other format—" and "None of the Anthropic equity held by the LLC will be used for lobbying or political giving, in the US or abroad." — same

With founder pledges and the employee donation match, "roughly 20% of Anthropic shares will be dedicated to the public good,".

Why it matters: a large new funder announced structure first, with eligibility and programme areas still unstated.

Corners of the web

A liver-organoid toxicity test brings ring-trial numbers

(translated from Korean) "Evaluated on 50 drugs by this criterion, sensitivity was 71.4%, specificity 100%, accuracy 80%." Original: "이 기준으로 50종 약물을 평가한 결과 민감도는 71.4%, 특이도는 100%, 정확도는 80%였다." — Yakup

(translated from Korean) "Within-laboratory reproducibility was 72–86%, between-laboratory reproducibility 82–96%." Original: "실험실 내 재현성은 72~86%, 실험실 간 재현성은 82~96%였다." — same

Four Korean institutions ran the validation under the OECD guidance document for new toxicity test methods.

Why it matters: this is the price of replacing an animal test, stated in public: no false positives, and three hepatotoxic drugs in ten missed.

People and good examples

Nancy Cox, who made sequencing the first measurement in influenza surveillance

"In 1992, Nancy became chief of CDC's Influenza Branch and Director of the WHO Collaborating Center" — Jernigan, Katz and Fukuda, Emerging Infectious Diseases, CC BY

"She promoted a "Seq-First" approach, in which genomic sequencing precedes traditional antigenic and serologic testing" — same

the network "Now comprising more than 160 National Influenza Centers in 136 countries and 7 Collaborating Centers" — same

The World Health Organization's own notice says she was "one of the principal architects of the GISAID Initiative" and that her leadership "shaped global influenza surveillance and response for nearly four decades" — WHO. She died in April, aged 77.

Why it matters: the measuring network, the shared sequence database and the tool that ranks new viruses for pandemic potential were built by the same person, inside a public agency, with sharing invited rather than compelled.

Cheruvayal Raman, who kept a rice collection alive by planting it

"He conserved more than 55 traditional rice varieties, safeguarding seeds that might otherwise have disappeared" and "He believed seeds should be shared, not sold" — South First

Seven years ago he described the reason and the terms in his own words: "our native paddy varieties were losing out to hybrid and genetically engineered seeds."; "I have never sold any of these seeds for any personal gain."; "I have not received any kind of support nor any grants from the government." — The Better India, 2019. Borrowers were asked to return an equivalent quantity after their next harvest. He died on 14 October in Wayanad, Kerala, at 74.

Why it matters: a reference collection maintained by cultivation rather than refrigeration, with a lending rule that turned every borrower into a backup site, and no documented succession.

Reads and ideas

Autonomous clinical AI by 2030, and six medical societies answering

A bioethicist and two co-authors argue in a journal viewpoint that autonomous systems "will likely be ready to be deployed for real-world cognitive medical tasks in some, maybe many, workflows by 2030." and that the open work is liability and payment, with the warning that "the more likely what docs do is going to be error-prone, rather than finding an AI mistake." — Chief Healthcare Executive.

Six organisations of family physicians, paediatricians, obstetricians, internists, surgeons and the American Medical Association replied that "Statements suggesting AI is inherently better informed than physicians" diminish expertise, that "Physicians evaluate patients in context, drawing on years of training and experience." and that "Pitting physicians against technology or suggesting AI should replace physician expertise" does not advance the work — ACP.

Why it matters: the disagreement is about where the cognitive work sits in a health system, and both sides locate the remaining obstacles outside the model.

The arithmetic of removing a funder, from someone who ran it

"I had 2,500 staff in sixty-five countries working to advance global health development," and "On a budget of $24 per American taxpayer—out of the $15,000 the average American taxpayer pays—they saved lives," — Gawande, Weatherhead Center

"These deaths unfold slowly." and "Further, the deaths are scattered, and that makes it hard to see." — same

He gives the reach figure as "instead of eighty-one million people, it's forty-three million people that are able to be reached."

Why it matters: a periphery that depends on one funder closes in a week, and the resulting deaths are distributed thinly enough that no routine statistic registers them.

Book of the day

Seeing Like a State, James C. Scott, 1998

"Certain forms of knowledge and control require a narrowing of vision." … "This very simplification, in turn, makes the phenomenon at the center of the field of vision" "more legible and hence more susceptible to careful measurement and calculation." — full text

"They did not successfully represent the actual activity of the society they depicted, nor were they intended to;" "they represented only that slice of it that interested the official observer." — same

Scott's subject is legibility: permanent surnames, standard weights, cadastral surveys, the forest reduced to board feet. His complement to legibility is mētis, the practical knowledge that resists being written down, and his example of it is a physician who recognised a disease early without being able to say how, until someone noticed he was reading a slight tremor of the eye.

The standing objection, from Scott Alexander's review, is that the book concedes most of modernity's gains and leaves no procedure for telling the wise local from the confident crank: "Well, for one thing, Scott basically admits to stacking the dice against High Modernism and legibility." (Slate Star Codex).

Why it matters: anyone building a register of a population, a disease or a literature is choosing what the schema is for, and Scott's question for the designer is what slice of reality it was built to see.