May 2026 — Monthly Briefing¶
1. What was read¶
865 research articles from 33 of the 33 journals on the roster.
A further 291 records (25% of the 1,156 this month's harvest returned) carry no abstract and are not research: news, correspondence, editor highlights and obituaries. PubMed types them Journal Article and nothing else, so the publication-type filter cannot see them — across the whole index they are 22.7% of it and 35% of general, and 3,801 of them are JAMA. Every count below is over the full harvest and is inflated by this; the figure above is not.
| General | Cancer | Immune | Bioinfo | Sysbio |
|---|---|---|---|---|
| 516 | 208 | 89 | 312 | 31 |
Volume swing against the median month: sysbio +29%
Check the source before reading anything into this. Two benign causes account for most of these. bioinfo every January carries the Nucleic Acids Research database issue and runs high. And several journals simply deposit nothing in some months: across the 44 months read, 4 of 33 journals are absent from at least one, worst being Nat Mach Intell (20 months missing, 81 papers in total) and Bioinformatics (4 months missing, 2,905 papers in total). Every journal on the roster contributed this month. Queried directly, PubMed returns zero for those journal-months too — the holes are in the source, reproduced exactly by re-running the harvest. So a swing here is not evidence that a field went quiet, and is not usually evidence that the pipeline broke either.
Largest contributors: Nature 123, JAMA 115, Nucleic Acids Res 94, Science 84, Bioinformatics 74, Lancet 64.
2. Net change on the thread curve¶
| Thread | This month | Cumulative | Status |
|---|---|---|---|
agentic-ai |
+2 / −1 | +32 / −9 | forming |
cell-therapy |
+4 / −2 | +232 / −75 | forming |
ctdna-mrd |
+3 / −4 | +83 / −29 | forming |
foundation-models |
+4 / −10 | +162 / −113 | splitting |
immunotherapy-resistance |
+8 / −5 | +258 / −95 | forming |
interception |
+9 / −1 | +157 / −27 | forming |
method-credibility |
−13 | +22 / −551 | crisis |
neoantigen-vaccines |
+1 / −4 | +84 / −50 | splitting |
neuro-immune-microbiome-axis |
+9 / −1 | +222 / −23 | forming |
next-gen-checkpoints |
−3 | +76 / −59 | splitting |
single-cell-3d-genome |
+3 / −2 | +89 / −23 | forming |
suppressive-population-inversion |
+1 / −2 | +56 / −105 | contested |
Status is derived by rule, not asserted. Fewer than 5 cards → too-early; then refutes ≥ 2×supports → crisis; refutes > supports → contested; supports ≥ 2×refutes → forming; else splitting.
3. The evidence that moved it¶
The heaviest month for foundation-model refutations the corpus had recorded to this point, and the form has changed. Ten cards, and three of the five read here replace the model with a trivial baseline and watch the number stay put: classical molecular fingerprints with shallow classifiers match or beat protein language models on peptide function; ML and deep survival models fail to beat Cox baselines on low-dimensional clinical data; and structure-aware substitution models built on predicted structures do not improve on plain sequence models. scArchon is the fourth and is more mixed than it was carded: across nine perturbation-response predictors on six single-cell datasets, performance is heterogeneous — trVAE, scGen, scPRAM and scVIDR are robust across multiple datasets, while others occasionally fall below linear or control baselines, and models with favourable aggregate scores can still lose the biological perturbation signature, which is the finding that matters. The fifth, 42020757 (Nature), is about the scoring rather than the model: accuracy-scored LLM benchmarks reward confident fabrication rather than measuring truthfulness.
Corrected 2026-09-04
The heading read "The strongest month for foundation-model refutations in the
corpus" and the count read "Five cards". Both fail. The 2026-09-04 abstract
re-scan puts May at ten refutations on foundation-models, matching the
−10 in section 2, and two later months exceed it: 2026-08 with twelve and
2026-07 with eleven. Ten was the maximum through May 2026 and the heading now
says only that. The five papers named above are unaffected — every one is
still carded refutes — and the observation the paragraph exists to make,
that the refutations have shifted from auditing a model to substituting a
trivial baseline for it, does not depend on the month being first.
Corrected 2026-08-27
This section carried a paragraph headed "ctDNA took two more hits, one of them a landmark". One of the two was not a hit. TOMBOLA was carded from a title that gives the trial, the intervention and the population and no endpoint, and a failure was supplied for it. The trial met its primary endpoint. The Galleri item is a news report rather than a trial paper, and it was credited with mortality evidence the trial never collected. Both are restated below.
TOMBOLA is the strongest single result the ctDNA-as-guide claim has. 41544704 (Ann Oncol) monitored 178 assessable patients after neoadjuvant chemotherapy and radical cystectomy for muscle-invasive bladder cancer with tumour-informed ctDNA, and started atezolizumab on ctDNA positivity irrespective of imaging. It met its primary endpoint: 60% (50 of 84) of treated ctDNA-positive patients reached molecular and radiographic complete response. Median lead time from ctDNA detection to imaging-confirmed recurrence was 90 days, and one-year recurrence-free survival was 97% in ctDNA-negative against 76% in ctDNA-positive patients. It is single-arm phase II, so it cannot say what would have happened without the atezolizumab — but ctDNA is the treatment trigger here, not a prognostic label, and that is the thread's claim.
The Galleri item is trade reporting, and its point is about endpoints. 41817436 is a Cancer Discovery news report, not a trial readout: NHS-Galleri, the randomised trial of the multi-cancer early detection test, missed its primary endpoint of a reduction in late-stage cancers, and researchers quoted in the piece attribute that partly to the trial's use of a surrogate endpoint rather than cancer-specific mortality. Mortality was never measured, so no mortality evidence was withheld or delivered; the criticism is of the design, and it is the same surrogate-endpoint problem this corpus tracks on the treatment side.
And the month's real negative for the thread is the one that worked. 41616224 (JCO) applied a tissue-free ctDNA assay to 2,260 patients from Alliance N0147 and found strong independent prognostication — disease-free survival hazard ratio 5.03, five-year DFS 27.7% against 77.1%, with tumour fraction stratifying further inside the positive group. The assay performs. It counts against this thread because the thread's claim is that ctDNA is more than prognostic, and here a large, well-executed application stops at risk stratification. Taken together the month reads: ctDNA detects reliably and early, a prospective trial has now acted on the detection and met its endpoint, and the month's largest cohort by an order of magnitude still uses it only to sort patients.
Anti-TIGIT got a defence rather than a result. 41564381
(JCO) argues that the original anti-TIGIT antibodies failed on Fc isotype
design — unmutated IgG1 — and that the target was therefore never properly
tested. The record in this index is a commentary title with no abstract, so the
argument is all there is to read, and it is an argument that withholds judgement
on the target rather than delivering one: the card is neutral. Note its
structure. The same manoeuvre — the target survives by relocating the blame to
the format — has now been applied to CD47, TIGIT and CTLA-4 in this corpus, and
it is the kind of claim that decides whether next-gen-checkpoints gets
reopened rather than the kind that settles it.
Interception found a second causal variable, alongside the mutation. 41781610 (Nature) shows that the stress response of nascent squamous tumours in the mouse upper gastrointestinal tract instructs the underlying mesenchyme to build a precancerous niche: stimulated fibroblasts mount a wound-healing response that remodels the extracellular matrix into a fibronectin-rich scaffold, and heterotypic 3D culture and grafting show that the niche alone is sufficient to confer tumour properties on carcinogen-free healthy epithelium. The paper's own model is that mutations and the stromal response to them together define whether an early lesion persists — not the niche instead of the clone. Read with February's finding that premalignant colorectal lesions are polyclonal in origin, the direction of 2026 is that the unit of interception is larger than a single mutant cell.
4. Refutations¶
(a) Landing on a tracked thread¶
41701973 (Blood): across 584 patients given CD19 CAR-T for large B-cell lymphoma at six Australian centres, time of infusion is not associated with outcome once confounders are accounted for. It is a clean null on a chrono-immunology hypothesis that a 2025 card had supported for allogeneic transplant, and it is recorded because the corpus should not accumulate only the positive half of a fashionable idea. It is not, though, a card against cell therapy: infusion timing is incidental to this thread's claim about generalisation beyond CD19-positive B-cell disease, and counting a null on an incidental variable as a refutation of the platform is how a thread acquires a negative it never earned. The card is now neutral.
(b) Method credibility¶
Fifteen cards, thirteen of them refutations and two now neutral. Of the four read here, only two are measurements. 41882360 (Nature) uses negatively supercoiled DNA minicircles and cryo-EM to show why supercoiling drives Cas9 off-target cleavage: the HNH domain is held in a more catalytically competent conformation, new DNA-RNA mismatch geometries are accommodated across the protospacer, and plasticity in the PAM-distal region is topology dependent. What the card previously added — that standard specificity assays cannot see this — is a methodological verdict the paper does not state, and it has been removed. It is the third editing-safety card in seven months, after November's heritable impairment after repair and December's epigenetic disruption by double-strand breaks. 42141119 (Nat Biotechnol): DNA off-targeting confounds RNA-chromatin occupancy assays, making a body of published RNA-chromatin maps artifactual.
Corrected 2026-09-04
This opened "Four cards, and only two of them are measurements." Four was the
whole month on this thread when May was carded from titles; the 2026-09-04
abstract re-scan puts it at fifteen — thirteen refutations, which is the
−13 in section 2, plus the two cards earlier corrections moved to neutral.
The four discussed here are described correctly and the half-are-measurements
observation holds of them. What is withdrawn is the implication that it holds
of the month: eleven cards on this thread are not read in this briefing at
all, and nothing here licenses a claim about their composition.
The other two of the four are not. 42133765 (Science) is a policy piece arguing that genetic quality control for mouse lines should be modernised; its abstract is a single sentence and it reports no measurement. The claim carded from it — that unmonitored drift is a hidden artifact behind irreproducible phenotypes — is plausible and was supplied by the reader, not by the paper, and it is now neutral.
Corrected 2026-08-27
This subsection also read 42129548 as an independent finding that analysis results shift with software package choice and version, and called it "the second month in a row with that finding, after April's 41903535, now from a different journal and a different data modality". It is a Nature Methods research highlight, single author and no abstract, and what it summarises is 41903535 itself. There is one study here, not two in successive months, and no second journal or modality. The card is now neutral.
What survives that correction is still worth naming. The audit genre spent 2025 moving from the model to the pipeline to the instrument, and April's 41903535 moved it to the version number — the claim is no longer that a method is wrong, but that the same method on the same data gives a different answer depending on when you ran it. That is one paper, given a second airing as a research highlight in another journal, and it should be carried at the weight of one paper until something independent lands on it.
5. Open items¶
Nothing carried forward.
6. Meetings¶
1 meeting with a written vault this month. These are read and summarised rather than harvested as abstracts, so they do not appear in the counts below.
790 abstracts from 2 meetings.
| Meeting | Abstracts | With full text |
|---|---|---|
| ESMO Breast Cancer 2026 | 595 | 0 |
| CIMT 23rd Annual Meeting | 195 | 194 |
7. Regulatory¶
| Event | Count |
|---|---|
| trial-first-posted | 622 (of which 567 genuinely oncology) |
| fda-approval | 9 |
| ema-authorisation | 6 |
9 FDA actions — dated from each notice's own prose, not the table's posting date, which disagrees on 7 of 190 rows and once by enough to move an approval into the wrong month:
- 2026-05-01 — FDA approves vepdegestrant for ER-positive, HER2-negative, ESR1-mutated advanced or metastatic breast cancer
- 2026-05-08 — FDA approves zenocutuzumab-zbco for advanced, unresectable or metastatic cholangiocarcinoma
- 2026-05-13 — FDA approves oral combination of decitabine and cedazuridine tablets with venetoclax for newly diagnosed acute myeloid leukemia
- 2026-05-13 — FDA grants accelerated approval to sonrotoclax for relapsed or refractory mantle cell lymphoma
- 2026-05-15 — FDA approves two separate indications for fam-trastuzumab deruxtecan-nxki in HER2-positive early-stage breast cancer
- 2026-05-15 — FDA approves atezolizumab for adjuvant treatment of muscle invasive bladder cancer in patients with molecular residual disease
- 2026-05-22 — FDA approves datopotamab deruxtecan-dlnk for unresectable or metastatic triple-negative breast cancer
- 2026-05-27 — FDA approves pivekimab sunirine-pvzy for blastic plasmacytoid dendritic cell neoplasm, an ultra-rare hematologic malignancy
- 2026-05-28 — FDA approves durvalumab in combination with Bacillus Calmette-Guerin for high-risk non-muscle invasive bladder cancer
3 EMA oncology authorisations:
- 2026-05-28 — Adstiladrin
- 2026-05-29 — Imdylltra
- 2026-05-29 — Zepzelca
8. Trade press¶
41 on-topic of 321 items across 3 feeds (n-stat 309, n-aacr-blog 9, n-acir 3). Headlines are selected by subject keyword, not recency — most of what these feeds carry is health policy and industry finance the corpus does not read.
- 2026-05-31 — Ultra-low doses could bring costly cancer treatments to more patients in poorer countries —
n-stat - 2026-05-31 — STAT+: For prostate cancer patients set on surgery, new hormone regimen may improve outcomes, study finds —
n-stat - 2026-05-31 — STAT+: Akeso and Summit’s ivonescimab extends survival in squamous cell lung cancer —
n-stat - 2026-05-31 — STAT+: Practice-changing results reported for Revolution Medicines pancreatic cancer drug —
n-stat - 2026-05-30 — STAT+: BioNTech and Pfizer tout bispecifics data at ASCO, and Pazdur sees silver lining at FDA —
n-stat - 2026-05-30 — STAT+: Revolution Medicines starts shipping experimental pancreatic cancer drug —
n-stat - 2026-05-29 — STAT+: At ASCO, positive data for Bristol in multiple myeloma and Pfizer in lung cancer —
n-stat - 2026-05-29 — STAT+: Pharmalittle: We’re reading about Replimune drug getting third try at FDA approval, a Pfizer deal in China, and more —
n-stat - 2026-05-28 — Editors’ Picks, May 2026: Insights Into CAR T Therapy, Neuron-Tumor Crosstalk in Pancreatic Cancer, and More —
n-aacr-blog - 2026-05-28 — STAT+: Ahead of ASCO, all eyes on pancreatic cancer —
n-stat - 2026-05-28 — STAT+: An ASCO preview: What to watch for at cancer research’s big meeting —
n-stat - 2026-05-27 — From Pixels to Patients: How AACR on Campus Is Transforming Cancer Research in Colombia —
n-aacr-blog - … and 29 more on-topic items.