← kaysauer.com ↑ Contents
BioMed & MedTech · week of 7 September 2026

What the Predictions Are Worth

Week 37 — three focus stories, moving outward from the model. Nine billion variant predictions with almost no ground truth; the one genome where a complete answer key exists; and three clinical deployments where the algorithm turned out not to be the hard part.
This Week's Overview

Last week ended on an open question: researchers had mutated nearly every base of bacteriophage ΦX174, the first genome ever sequenced, and the AI models built to predict exactly that kind of effect struggled. This week follows that question outward, away from the model and toward the patient, and the pattern that emerges is consistent. DeepMind released an atlas scoring all nine billion possible single-nucleotide changes in the human genome, with a second genome-scale model landing days later. Then we ran a DNA language model on the ΦX174 benchmark itself — same rows, same metric, against the same 34 protein predictors.

Prediction gets genome-scale, and gets scored

The benchmark result is more interesting than either a win or a loss. Evo2, which has never seen a protein sequence or a multiple-sequence alignment, ranks 16th of 35 — ahead of every ESM-family model in the comparison, behind the alignment- and structure-aware specialists. It also produces something no protein model can: a score at the nucleotide level, in a genome where genes overlap so densely that three of them have no unambiguous positions at all. That is the honest shape of the prediction side right now. Coverage is enormous, validation is tiny, and the only place the two properly meet is a 5,386-base phage.

And then what?

Which makes the third story the one that decides whether any of it matters. Three clinical deployments reported this week — a decision-support tool in lung cancer, a randomized mammography-screening trial, an agentic eye clinic in Beijing — and in all three the limiting factor was not model accuracy. It was where the output sits in the workflow, whether the clinician clicks through to it, and whether anyone chose an endpoint that tracks patient benefit rather than algorithmic skill. One of them recovered its physician usage rate from 3.8% to 23% by cutting the interaction down to three clicks. Below the three focus stories, Also This Week collects the rest: what cancer treatment does to healthy tissue, four new accounts of where tumours hide, a drug-delivery collection that has moved from organ to organelle, and genomes read end to end.

Contents

Focus Topics

  1. An atlas of every mutation you could possibly make
  2. A DNA model walks into a protein benchmark
  3. Did the AI change what the doctor did?

Also This Week

  1. What the treatment does to the healthy tissue
  2. Where tumours hide from the immune system
  3. Drug delivery moves from the organ to the organelle
  4. Genomes read end to end

Test Yourself

Read the newsletter and you'll be able to answer these.
  1. AlphaGenome scores about nine billion variants. What does it actually predict for each one, and why is that not the same as predicting pathogenicity?
    Show answer
    It predicts molecular consequences — effects on gene expression, splicing and chromatin — not clinical outcome. A variant can measurably shift expression without causing disease, and can cause disease through a mechanism the model does not represent. The atlas is a layer of molecular annotation that still has to be connected to phenotype by something else.
  2. What is the central difficulty in validating a variant-effect model at genome scale?
    Show answer
    There is no ground truth anywhere near that size. The largest complete experimental mutagenesis of an entire genome covers about 44,500 variants in a 5,386-base phage. Nine billion predictions against tens of thousands of measurements is roughly five orders of magnitude of coverage with no corresponding measurement — so most of the atlas can only be checked indirectly, against conservation, known clinical variants, or other models.
  3. GPN-Star and AlphaGenome both score variants genome-wide. What is the architectural difference in what they read?
    Show answer
    GPN-Star is phylogeny-aware: it reads whole-genome multiple-alignment data across species, so its notion of “this position matters” is grounded in evolutionary constraint. AlphaGenome predicts functional readouts (expression, splicing, chromatin) from sequence. One is essentially a learned, flexible replacement for conservation scores; the other is a predicted molecular phenotype.
  4. Evo2 was scored zero-shot on the ΦX174 benchmark. What does “zero-shot” mean precisely here, and what does it not mean?
    Show answer
    It means the fitness measurements were never a training target — the same situation every protein predictor in the comparison faces. It does not mean the genome was unseen: ΦX174 was the first DNA genome ever sequenced and is almost certainly somewhere in Evo2's 8.8-trillion-nucleotide pretraining corpus. So this measures scoring ability on a known genome, not generalization to a new one.
  5. How is a variant actually scored, mechanically?
    Show answer
    B. A bigger drop in the model's own likelihood means it finds that DNA change more disruptive. That delta is correlated against the lab-measured fitness effect, per gene, using Spearman correlation — matching the source paper's own evaluation exactly, which is what makes the ranking comparable.
  6. Where did Evo2 land, and which models did it beat?
    Show answer
    16th of 35 at the amino-acid level, mean per-gene Spearman ρ=0.257 (median 0.329) over 28,380 variants. It beat every ESM-family model tested — ESM1v, ESM2 up to 15B parameters, ESM3, ESMC up to 6B — plus Tranception and PoET. It trailed the MSA- and structure-aware specialists (ESCOTT, GEMME) and the strongest single-sequence protein language models (VESPA, Progen2).
  7. At the nucleotide level Evo2 scored ρ=0.55–0.65 for every gene with an unambiguous reading frame. Why were genes B, E and K excluded, and why is that not a flaw in the analysis?
    Show answer
    ΦX174's genome is so densely packed that those three genes are entirely contained within other genes' reading frames — every single base in them also codes for a different protein. So there is no nucleotide position whose effect can be attributed to one gene. That is a real property of the organism, not missing data.
  8. Two data problems had to be fixed before any score could be trusted. What were they?
    Show answer
    First, the lab's actual wild-type genome differs from the canonical NCBI reference at four fixed positions — a real strain difference — so the wild-type used for scoring was reconstructed from the fitness data itself rather than downloaded. Second, the source table's amino-acid residue numbering was 0-based for two of the eleven genes and 1-based for the other nine; the fix was to locate every substitution by diffing the mutant protein sequence against the correct gene translation instead of trusting the numbering column at all.
  9. The NSCLC decision-support study reported two separate results. What were they, and why is the second one unusual?
    Show answer
    The multimodal explainable-AI model outperformed established biomarkers at predicting immunotherapy outcome, and physicians using it made measurably better decisions. The second is unusual because most clinical-AI evaluations stop at model accuracy and never establish that anything downstream actually changed.
  10. MASAI used AI to reorganize mammography screening rather than to read it. How did the workflow actually work?
    Show answer
    The AI produced a risk score from 1 to 10, shown to radiologists before they interpreted the examination. High-risk examinations were double-read, lower-risk ones single-read. The AI's detection marks were deliberately withheld until the end of case reading, because eye-tracking work shows that visible marks pull readers into highlighted areas instead of a holistic search. The score was used to allocate human attention, not to make the call.
  11. Why did MASAI choose interval cancer rate as its primary outcome instead of cancer detection rate?
    Show answer
    B. An intervention that mainly finds low-grade in situ lesions or indolent cancers looks good on detection rate while doing little or nothing for patients. Interval cancers — those appearing after a negative screen but before the next round — generally carry worse prognosis and are widely treated as a proxy for breast-cancer mortality.
  12. What did MASAI find, and at what cost in reading workload?
    Show answer
    A 29% increase in cancer detection, driven predominantly by small, lymph-node-negative invasive cancers including a higher proportion of biologically aggressive subtypes, and 12% fewer interval cancers — with recall rate, false positive rate and specificity essentially unchanged. Screen-reading workload fell 44%. A pre-specified safety analysis at roughly 80,000 women had already confirmed detection did not decline before the trial proceeded to full evaluation.
  13. The Beijing eye-clinic deployment saw physician usage collapse and then recover. What happened?
    Show answer
    Usage peaked at 25.7% of examinations in December 2025, fell to 3.8% by April 2026, then returned to 23.0% in May 2026 after the team optimised the interface and loading speed, cut the interaction to three clicks and limited manual input. Nothing about the model changed. The authors' conclusion is that workflow usability is not a secondary implementation detail but a continuous determinant of whether the system produces any clinical value at all.
  14. In the same deployment, what raised the retinal model's macro-AUROC from 0.7996 to 0.9383?
    Show answer
    Data curation, not modelling. RETFound fine-tuned on 26,839 routinely collected fundus images performed poorly on the first prospective cohort. Four ophthalmologists manually selected 1,426 high-quality images with verified diagnostic labels, and fine-tuning on those produced the jump. A further round of prospective feedback took it to 0.9482.
  15. What is the “silent trial paradox,” and what is the “referral fallacy”?
    Show answer
    A silent trial deliberately prevents clinicians from seeing or acting on the AI's output — so it can only ever measure detection performance, never the thing the system exists to do. The referral fallacy is that successful detection is one step in a long pathway: real outcomes depend on referral acceptance, specialist access, workup and treatment delivery, so accurate screening alone may change nothing if the downstream process is broken.

From Last Week's Issue — 2026-09-05

  1. The old view was that microglia are laid down before birth and self-renew for life. Which single measurement technique broke that picture, and why did gene-expression studies miss it entirely?
    Show answer
    DNA methylation, profiled cell-by-cell. Methylation marks preserve a cell's lineage of origin, so a monocyte-derived cell that has moved into the brain and started behaving like a microglia still carries a blood signature in its methylome. Gene expression is set by the current environment, so those cells look exactly like microglia by transcription alone.
  2. CHIP — clonal haematopoiesis — is normally treated as a risk factor. What did last week's studies find about CHIP and Alzheimer's?
    Show answer
    The opposite of the usual story: people with larger CHIP clones had substantially lower Alzheimer's risk and less amyloid and tau. The leading guess is that expanded monocyte clones are better at phagocytosis — clearing amyloid and tau — once they become brain-resident.
  3. How does daraxonrasib inhibit RAS, and how is that different from the KRAS-G12C drugs before it?
    Show answer
    It is a molecular glue: it first binds cyclophilin A, and that complex grips both normal and mutant RAS in its active state, blocking RAS from engaging downstream partners. KRAS-G12C inhibitors target one specific mutant of one RAS gene; daraxonrasib hits the pathway broadly.
  4. When researchers made every possible single change to ΦX174, what fraction of the harmful amino-acid mutations could they not explain?
    Show answer
    About a quarter. Roughly half of the harmful changes disrupted protein–protein interactions, another quarter hit buried structural residues, and the remaining quarter had no identifiable cause. This week's focus topic scores a DNA model on that same dataset.

An atlas of every mutation you could possibly make

Nature, 9 September 2026

DeepMind has released AlphaGenome, a model that predicts the functional consequences of every possible single-nucleotide change across the human genome — roughly nine billion of them. For each variant it estimates effects on gene expression, splicing and chromatin, and the whole thing is pitched as a reference atlas: a lookup table for interpreting the non-coding variation that dominates every association study and that clinical genetics has never had a good story for.

The scale is the headline, and it deserves a moment. Nine billion is not a sample of interesting positions. It is the complete space of single-letter changes you could make to a human genome, scored. Nothing in molecular biology has been annotated at that granularity before, and the non-coding fraction is exactly where the field has been stuck — a GWAS hit lands in a regulatory desert, and the honest answer to “which base is doing the work, and how?” has usually been that nobody knows. It is worth reading alongside a Nature Genetics review published a day earlier, which works through the mechanisms by which small non-coding variants in promoters, enhancers and silencers actually cause monogenic disease — a reminder that “affects expression” covers a lot of distinct causal routes, and that systematically discovering such variants remains an open methodological problem rather than a solved lookup.

It did not arrive alone. The same week, “Predicting genome-wide functional constraints with GPN-Star” describes a genomic language model with a phylogeny-aware architecture that reads whole-genome multiple-alignment data, estimating functional constraint and variant effects across the entire genome and across species. It is positioned as a more scalable and flexible replacement for conservation scores. The two models are doing different things — GPN-Star grounds “this position matters” in evolutionary constraint, AlphaGenome predicts a molecular readout from sequence — but they are aimed at the same clinical bottleneck, and they arrived within days of each other.

Which raises the question this issue keeps circling. A predicted effect on expression is not a prediction of disease. A variant can shift expression measurably and cause nothing; it can cause plenty through a mechanism the model does not represent. And there is no ground truth at anything like nine billion. The largest complete experimental mutagenesis of an entire genome that exists covers about 44,500 variants — in a 5,386-base bacteriophage. That is the gap.

PREDICTED vs. MEASURED — LOG SCALE Variants scored by AlphaGenome 9,000,000,000 Variants with a measured fitness effect (ΦX174, whole genome) 44,478 Bars are log-scaled. On a linear axis the orange bar would be about 1/200,000 of a pixel.
Own diagram. Counts from the Nature news piece and the ΦX174 complete-mutagenesis dataset.
Why it matters

Genome-scale variant scoring is now real and there are at least two credible implementations of it. What has not scaled is the measurement side, and that asymmetry decides how these atlases should be used: as a prioritisation layer that narrows where to look, not as an answer. The test case is the next story — the one genome where a complete measured answer key exists.

A DNA model walks into a protein benchmark

Own work, completed 11 September 2026 — benchmark dataset from Wei, Li & Lehner, bioRxiv 2026 (CC BY)

Last week's issue covered the ΦX174 result: Wei, Li and Lehner measured the fitness effect of essentially every possible mutation in the phage's genome — 16,098 single-nucleotide and 28,380 single-amino-acid substitutions, 44,478 in total — and benchmarked roughly fifteen variant-effect predictors against the real numbers. Every one of those predictors reads protein sequence, or a multiple-sequence alignment built from related proteins. Not one of them looks at DNA.

That is a conspicuous hole, and it is the one this project fills. Evo2 is a genome-scale DNA language model from the Arc Institute and NVIDIA, StripedHyena 2 architecture, pretrained on 8.8 trillion nucleotides across bacteria, archaea and phage. Scoring it is mechanically simple: for each of the 44,478 designed mutations, take the model's log-likelihood of the mutant genome and subtract its log-likelihood of the wild type. A bigger drop means the model finds that specific change more disruptive. Correlate that delta against measured fitness, gene by gene, using the paper's own per-gene Spearman metric — same rows, same statistic, so the number drops straight into the published ranking table.

At the amino-acid level, Evo2 ranks 16th of 35, mean per-gene ρ=0.257, median 0.329, across 28,380 variants. The interesting part is who sits on either side of it. A model that has never been shown a protein sequence or an alignment beats every ESM-family model in the comparison — ESM1v, ESM2 up to 15B parameters, ESM3, ESMC up to 6B — along with Tranception and PoET. It trails the specialists that get to use extra information: the MSA- and structure-aware methods (ESCOTT, GEMME) and the strongest single-sequence protein language models (VESPA, Progen2).

AMINO-ACID BENCHMARK — RANK OF 35 PREDICTORS 1 35 EVO2 — rank 16, ρ=0.257 reads DNA only Ahead: ESCOTT, GEMME (alignment / structure), VESPA, Progen2 (single-sequence protein LMs) Behind: every ESM-family model tested (ESM1v, ESM2 15B, ESM3, ESMC 6B), Tranception, PoET Positions shown schematically; only Evo2's own rank and correlation were computed here.
Own diagram. Rank and ρ from the zero-shot run; the surrounding predictors are those named in the source paper's table.

The nucleotide-level result has no precedent to compare against, because no protein model can be scored on raw DNA at all. Across every gene with an unambiguous reading frame, Evo2's predictions correlate with measured fitness at ρ=0.55–0.65, all p<0.001. Genes B, E and K are absent from that list, and the reason is a property of the organism rather than the analysis: ΦX174's genome is packed so tightly that every base in those three genes also sits inside another gene's reading frame, so no position can be attributed to one gene alone.

Two things had to be fixed before any of it could be trusted, and both are the kind that quietly corrupt a result. The lab's actual wild-type genome differs from the canonical NCBI reference at four fixed positions — a genuine strain difference — so the wild type used for scoring was reconstructed from the fitness data itself rather than downloaded. And the source table's residue numbering turned out to be 0-based for two of the eleven genes and 1-based for the other nine; rather than trust the numbering column, every substitution was located by diffing the mutant protein sequence against the correct gene translation. The compute was less elegant: a Colab A100 preempted about nine times over roughly a day of wall-clock GPU time, survived by checkpointing scored variants to disk every two and a half minutes.

The caveat belongs attached to every number above. This is zero-shot in the sense that the fitness measurements were never a training target, which is the same position every protein model in the comparison is in. But ΦX174 was the first DNA genome ever sequenced, and it is almost certainly somewhere in Evo2's pretraining corpus. This is a test of scoring on a known genome, not of generalisation to an unseen one.

Why it matters

It is a concrete data point for the gap the atlas story leaves open. Reading DNA directly, with no protein representation and no alignment, is already competitive with mid-tier protein-specific methods — which suggests the sequence itself carries more of the signal than the field's architecture choices assume. It also shows what an honest benchmark costs: a known answer key, the original paper's exact metric, and a willingness to report 16th place.

Did the AI change what the doctor did?

Nature Medicine, 7–13 September 2026 expanded from full text

Three clinical AI deployments reported this week, in three specialties, on three continents. What makes them worth reading together is that none of them is really about model accuracy — and two of them say so outright.

Start with the one that looks most like a conventional result. A large international real-world study in non-small cell lung cancer tested a multimodal explainable-AI decision-support tool and reported two things: the model outperformed established biomarkers at predicting immunotherapy outcome, and physicians using it made measurably better decisions. The second finding is the rare one. Most clinical-AI papers establish the first and quietly assume the second follows. The accompanying News & Views is more restrained, noting that pooling many patient-level biomarkers across modalities reliably beats any single one, but that the models degrade across sites — leaving generalisability, not accuracy, as the open problem.

The deepest account of why that gap exists comes from Kristina Lång, who spent nearly a decade on AI in mammography before running the MASAI randomized trial. Her retrospective work had been encouraging — very few screen-detected cancers got low AI risk scores, and a substantial share of interval cancers turned out to be visible in hindsight and correctly flagged. But, as she puts it, retrospective studies “do not tell us whether radiologists will trust it, ignore it, over-rely on it or adapt their behavior in unexpected ways.”

So MASAI used the AI to reorganise the work rather than to do it. The system produced a risk score from 1 to 10, and that score was shown to radiologists before they read the examination: high-risk cases were double-read, lower-risk cases single-read. The reasoning is unusually candid about human factors. Radiologists already carry a bias — they know cancer is rare in a screening population, and rare targets get missed more often — so rather than trying to eliminate that effect, Lång tried to aim it, raising suspicion where disease probability was high and lowering it where probability was low. A second choice went the same way: the AI's detection marks were deliberately withheld until the end of case reading, because eye-tracking work shows that visible marks pull readers toward highlighted regions instead of a holistic search. “This was a behavioral decision rather than a technical one.”

The endpoint choice mattered just as much. Detection rate is the obvious metric and a misleading one: an intervention that preferentially finds indolent lesions scores well while helping nobody, and earlier detection can inflate survival through lead-time bias without moving mortality. MASAI used interval cancer rate instead — cancers appearing after a negative screen but before the next round, which carry worse prognosis and serve as a recognised proxy for screening efficacy. The trial ran as a non-inferiority design with a pre-specified safety analysis at roughly 80,000 women, which found detection had not declined despite a 44% cut in screen-reading workload. The completed trial then reported a 29% increase in cancer detection, driven predominantly by small, node-negative invasive cancers including a higher share of aggressive subtypes, and 12% fewer interval cancers — with recall rate, false positive rate and specificity essentially unchanged.

The Beijing deployment supplies the unglamorous counterpart. AI-TEC is not one model but a coordinated set of agents spanning the whole ophthalmic pathway — pre-consultation, triage, diagnosis, clinician decision support, patient follow-up — deployed into the hospital information system at Beijing Tsinghua Changgung Hospital in November 2025. Its authors argue that bolting AI onto an existing workflow tends to create “an additional information layer that clinicians must now manually review and integrate, creating duplicated effort rather than reducing workload.” Then their own usage numbers made the point for them.

SHARE OF EXAMINATIONS WHERE A PHYSICIAN USED THE SYSTEM 30% 0 25.7% Dec 2025 206 / 801 3.8% Apr 2026 41 / 1,113 23.0% May 2026 259 / 1,126 interface fixed: 3 clicks Three reported months; the dashed segment is a decline, not measured monthly. The model was unchanged throughout.
Own diagram, from the usage figures reported in the AI-TEC implementation report.

Usage peaked at 25.7% of fundus examinations in December 2025, slid to 3.8% by April 2026, and came back to 23.0% in May after the team sped up loading, trimmed the interaction to three clicks and cut manual input. The model did not change. Their data story runs the same way: RETFound fine-tuned on 26,839 routinely collected images did poorly on the first prospective cohort, and what fixed it was four ophthalmologists hand-selecting 1,426 high-quality images with verified labels, which lifted macro-AUROC from 0.7996 to 0.9383; a further prospective feedback round reached 0.9482. They also name four traps in how this work is normally evaluated: the “silent trial paradox” — trials that prevent clinicians from acting on the AI can only measure detection, never care; the assumption that a hospital diagnosis is ground truth when specialists routinely disagree; the mismatch between disease-centric models and symptom-centric medicine, where the clinician's question is why this person came in; and the “referral fallacy,” that detection is one step in a pathway whose later stages decide whether anything improves.

The week's fourth item is the control case. ProteinTalks, a virtual-cell model built on proteomics rather than single-cell sequencing, was trained on more than 38 million protein measurements from 18 breast-cancer cell lines treated with 63 approved drugs and 59 combinations, sampled before treatment and at 6, 24 and 48 hours. It predicts cellular response to 81 held-out drugs with 88% accuracy, and in biopsies from 501 patients it predicted clinical outcomes of the treatments actually given. For three patients it screened 3,000 compounds and recommended the drugs they had received — plus three with better efficacy in the lab. Impressive numbers, and co-author Tiannan Guo is the one insisting it remains a prototype pending prospective trials. Which is precisely the step the other three papers are about. Meanwhile the Validation Accords is recruiting collaborators for a consensus framework on validating generative AI in medicine, on the observation that real-world validation is frequently skipped altogether, with external validation treated as the ultimate test.

Why it matters

Across all three deployments the binding constraint sits downstream of the model: endpoint choice in MASAI, click count in Beijing, site-to-site generalisation in the lung-cancer study. That is the same lesson the atlas and the benchmark arrive at from the other end — a score is not a finding, and a finding is not a changed decision. Lång's version is the most quotable: the future of medical AI “may depend less on building smarter algorithms than on designing better collaborations between humans and machines.”

Also This Week

What the treatment does to the healthy tissue

Two independent lines of work this week ask what cancer therapy leaves behind in tissue that was never cancerous. Duplex sequencing of normal tissues from childhood-cancer patients shows platinum chemotherapy drives paediatric mutation burdens up to adult levels and leaves a liver-specific mutational signature not seen in other tissues, generating nonsynonymous variants — some potentially leukaemogenic — that may help explain late effects in survivors; an accompanying Perspective reads the damage as accelerated ageing of the organ. From the other direction, a Nature Genetics study of normal oesophagus finds that short-course chemo and radiotherapy before surgery selects pre-existing mutant clones — in DNA damage response, HLA and kinase genes — without leaving any detectable mutational signature at all, meaning the usual signature-based forensics would miss it entirely. The Research Briefing notes that some of the selected mutants are druggable and confer treatment resilience on normal cells. Sequencing healthy tissue from treated patients, in other words, may reveal resistance genes that sequencing the tumour does not.

Where tumours hide from the immune system

Four papers, four different hiding places. PLA2G2D-high macrophages in tumour-draining lymph nodes track with poor prognosis and inhibiting the enzyme improves anti-tumour immunity in mice — the draining node, not the tumour, as the site of suppression. A serpin–myeloid axis in pancreatic cancer builds fibrin-rich niches that reprogram macrophages locally and physically exclude T cells, producing spatially organised pockets of evasion rather than a uniformly cold tumour. Chronic type II interferon resolves a long-standing paradox: sustained exposure releases double-stranded mitochondrial RNA into the cytoplasm, triggering a type I interferon response that synergises with IFN-II through prostaglandin synthesis to accelerate tumour growth — the same cytokine, opposite sign, depending on duration. And proximity-guided graph learning defines “proximity antigens” — pairs of surface proteins that sit close together on tumour cells but not healthy ones — identifying EGFR–CDCP1 as a co-target pair for multispecific therapeutics, which turns spatial arrangement rather than expression level into the selectivity mechanism. Separately, a Perspective on PD-1 argues the checkpoint works by dissolving the biomolecular condensates that organise T cell receptor signalling, a mechanism distinct from the phosphatase-recruitment model underpinning current checkpoint-inhibitor thinking. The clinical counterpart to all of this is to intervene before the niche exists: in the randomised phase 2 ImmunoPRISM trial, the BCMA-directed bispecific teclistamab produced higher complete-response rates than lenalidomide–dexamethasone in high-risk smoldering myeloma — treating patients who do not yet have overt disease — though follow-up is not yet long enough to show durable prevention of progression.

Drug delivery moves from the organ to the organelle

Nature ran a drug-delivery Outlook this week, and the collection has a clear direction of travel. The framing piece argues that targeting precision now limits therapeutic benefit more than potency does — a good drug that misses is not a good drug. One feature covers the competing nanoparticle and exosome approaches to the blood–brain barrier, the obstacle that has stalled brain-cancer and neurodegeneration treatment for decades; another goes a level finer, to therapies aimed at mitochondria and other organelles rather than merely the right cell, several of which are now in clinical testing. The collection's least glamorous entry may be the most consequential: a Q&A with health psychologist Rob Horne on medication non-adherence — the largest single source of lost therapeutic benefit in medicine, and one that no amount of delivery engineering addresses.

Genomes read end to end

Complementing the prediction story, several papers this week are simply about reading sequence more completely. A telomere-to-telomere mouse Y chromosome closes all 25 gaps in the GRCm39 reference, corrects structural errors, adds 8.7 Mb of previously unassembled sequence and 142 new genes, maps pseudoautosomal recombination loci and turns up a Y-chromosome dosage-compensation mechanism from X–Y homologous gene expression. On the human side, the T2T-HG002v1.1 benchmark extends variant-calling ground truth into regions that were previously unassessable. And two population-genetics papers push on what ancestry actually explains: a within-family design in a Mexico City cohort — which controls for the population stratification that confounds standard ancestry comparisons — still finds significant ancestry-associated differences in height and type 2 diabetes, while a schizophrenia CNV study in East Asian populations identifies rare copy-number variants and, meta-analysed with European-ancestry data, additional risk loci enriched in loss-of-function-intolerant genes.