February 2026 — Monthly Briefing¶

1. What was read¶

891 research articles from 32 of the 33 journals on the roster.

A further 296 records (25% of the 1,187 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
498	181	94	387	27

Volume swing against the median month: bioinfo +43%

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). This month is missing Nat Mach Intell. 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: Nucleic Acids Res 140, JAMA 124, Nature 101, Bioinformatics 95, Science 89, Genome Biol 49.

2. Net change on the thread curve¶

Thread	This month	Cumulative	Status
`agentic-ai`	+4 / −2	+22 / −8	forming
`cell-therapy`	+2	+208 / −67	forming
`ctdna-mrd`	+2	+77 / −24	forming
`foundation-models`	+1 / −4	+140 / −93	splitting
`immunotherapy-resistance`	+2 / −1	+226 / −88	forming
`interception`	+3 / −1	+135 / −24	forming
`method-credibility`	+1 / −6	+21 / −501	crisis
`neoantigen-vaccines`	—	+81 / −44	splitting
`neuro-immune-microbiome-axis`	+3	+199 / −20	forming
`next-gen-checkpoints`	+2	+72 / −49	splitting
`single-cell-3d-genome`	+1	+79 / −18	forming
`suppressive-population-inversion`	—	+53 / −98	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¶

Corrected 2026-08-27

This section opened by claiming that the interception thread took three independent hits in one month, all on the same assumption. On re-reading the abstracts, one of the three was a second telling of another and one pointed the other way.

41528342 has no abstract and no named authors: it is a Cancer Discovery research-watch restatement of 41291291, the Nature paper published five days earlier and carded separately on this thread. Counting both put one study on the interception curve twice; it is now neutral.

41380103 was read from its title. Its conclusion is that genomics separates progressor from nonprogressor smoldering myeloma, and it delivers a score, validated on 101 independent samples, that adds to 20/2/20. It is now carded supports.

One genuine hit survives, and the paragraph below states it.

Premalignant colorectal lesions are not one clone caught early. 41291291 (Nature) reports that human premalignant colorectal lesions are polyclonal in origin. Interception is built on the premise that a precursor is a single clone caught early, and this says the unit of the thing being intercepted is not what the field has been modelling. It is one paper, in one tissue, and it should be carried as one.

Precursor staging got sharper rather than looser. 41380103 (JCO) sequences 224 precursor samples, 51 of them paired with the myeloma the same patient went on to develop, against 1,779 newly diagnosed myelomas. Progressor smoldering myeloma is already genomically myeloma — over 80% of driver mutations present at both time points, most progressor samples showing only minor clonal change — while nonprogressor smoldering myeloma carries a lower mutational load, no chromosome 8 copy-number alterations and an MGUS-like profile. The genomic score built on that separation identifies low-risk patients unlikely to progress and adds clinical value on top of 20/2/20. The assertive title dissolves the old taxonomy; the result underneath it makes precursor staging more discriminating, not less, which is what interception needs.

Corrected 2026-08-27

The paragraph that follows read: "SERENA-6 is the month's most consequential negative, and it is a patient-reported outcome … ctDNA-guided early switching produces no patient-perceived quality-of-life gain." That is the opposite of what 41125211 reports. The card was written from the title, which names the trial, the drugs and the population and gives no result, and a null was supplied for it. Every reported hazard ratio in the abstract favours the switching arm. The argument built on it — a surrogate/endpoint gap visible inside the thread's own flagship — has no support here and is withdrawn.

SERENA-6 reported its patient-reported outcomes, and they favour the molecular trigger. 41125211 (Ann Oncol) reports that switching from aromatase inhibitor to camizestrant on emergent ESR1 mutation, with the CDK4/6 inhibitor continued, delays deterioration in pain (hazard ratio 0.57), dyspnoea (0.52) and emotional functioning (0.51), alongside the previously reported global health status hazard ratio of 0.54. SERENA-6 is the trial that made molecular relapse rather than clinical relapse the trigger for changing therapy, and this is the result that says patients report the molecularly triggered switch as better, not only the scan.

The neuro-immune axis produced two supports and no refutations, both mechanistic and both in top general journals: a tumour-brain sensory-sympathetic axis restraining antitumour immunity (41639447), and sensory neurons driving immune exclusion through dense extracellular matrix (41650969). The second is the more useful, because it supplies a physical mechanism — matrix density — for a thread that has mostly trafficked in signalling. A third card, 41629655, was counted here as serotonin driving chemoresistance; it is a Nature Immunology research highlight, single author, no abstract, and the study it summarises is not in this index. Nothing about that finding can be checked from the record, and it is now neutral.

4. Refutations¶

(a) Landing on a tracked thread¶

Withdrawn from this month 2026-08-27

This section carried a blinatumomab / CD19 CAR-T result attributed to 41643692 in Lancet Oncology. That PMID is "Time to act: a global call for oral cancer prevention and early detection" and has nothing to do with the claim.

The finding is real — it is 41643192, a single transposed digit away, in Blood. But it is dated 2026-04, so it was never a February result and cannot stay in this briefing. It has been re-pointed in data/evidence.yml and belongs to April.

Every automated check passed this for months: the PMID existed, its date matched the index, and its stored title matched the paper. What none of them test is whether the card is about the paper it cites.

Two cards this month are engineering around a failure rather than evidence for or against a target. 41735000 proposes depleting TIGIT-high CD226-negative clonal Tregs as a safer route to TIGIT targeting. 41734999 is a JITC commentary on work by Du et al., in which an Arg1-promoter-driven CD47 CAR macrophage regressed breast and gastric tumours in mice with minimal erythrocyte toxicity — secondary reporting of another group's preclinical mouse work, not primary data, and the comparison with the haematotoxicity that sank the CD47 antibodies is this corpus's inference rather than the commentary's result. Both are filed as supports on next-gen-checkpoints, but note what they are: the target survives only by being approached through a different cell or a narrower subset. That is the same manoeuvre as CD47 reappearing in December 2025 as a CAR-macrophage homing address rather than a drug.

(b) Method credibility¶

41559218 (Nat Genet) is the strongest card of the month: cell-segmentation error frequently dominates differential expression, neighbourhood and ligand-receptor results in imaging spatial transcriptomics. Read with January's 41519776, two consecutive months now say the two most common operations in imaging spatial transcriptomics — deciding where cells are, and testing which genes differ — are unreliable at the base.

41601205: reported transcription-factor binding-site accuracy is an artifact of how the negative set was constructed, not of model capacity. 41580962: drug-synergy deep models ignore chemical structure entirely, and one-hot drug identity reproduces their reported performance. These are the cleanest kind of refutation available — the model is replaced by a trivial baseline and the number does not move.

41736096 was carded here as evidence that published cfDNA-methylation cancer classifiers carry demographic bias inflating their apparent performance. It says something narrower and more constructive: demographic imbalance in training data degrades performance for minority populations, and DeBias removes the bias-associated subspaces from the feature space and improves cancer detection for those populations, with the gains reproduced in independent cohorts. It makes no claim about the performance of published classifiers, and it is now neutral. 41742311: a benchmark of LLM agents on single-cell omics quantifies where they break down on real workflows — the maturity signal for agentic-ai is the arrival of benchmarks, and it keeps arriving as bad news.

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.

AACR Immuno-Oncology Conference 2026

No abstract book was deposited this month either — the vault above is built from transcripts and programmes rather than from a published abstract set. The abstract layer runs 2023-02 to 2026-08 and is seasonal by nature — meetings are annual, so most months are empty here and that is the expected shape, not a gap in the harvest.

7. Regulatory¶

Event	Count
trial-first-posted	677 (of which 609 genuinely oncology)
ema-authorisation	7
fda-approval	5

5 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:

2026-02-05 — Safety labeling update for capecitabine and fluorouracil (5-FU) on risks associated with dihydropyrimidine dehydrogenase (DPD) deficiency
2026-02-10 — FDA approves pembrolizumab with paclitaxel for platinum-resistant epithelial ovarian, fallopian tube, or primary peritoneal carcinoma
2026-02-19 — FDA approves acalabrutinib with venetoclax for chronic lymphocytic leukemia or small lymphocytic lymphoma
2026-02-24 — FDA grants traditional approval to encorafenib for metastatic colorectal cancer with a BRAF V600E mutation
2026-02-26 — FDA grants accelerated approval to zongertinib for unresectable or metastatic non-squamous non-small cell lung cancer

2 EMA oncology authorisations:

2026-02-12 — Aumseqa
2026-02-16 — Anktiva

8. Trade press¶

37 on-topic of 300 items across 3 feeds (n-stat 287, n-aacr-blog 9, n-acir 4). 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.

2026-02-26 — STAT+: What stands between brain implants and FDA approval? — n-stat
2026-02-26 — STAT+: Brain implants: What’s standing in the way of pivotal trials, FDA approval — n-stat
2026-02-26 — Opinion: I tried to design a clinical trial for a supplement. 5 things got in the way — n-stat
2026-02-25 — Editors’ Picks, February 2026: FDA’s Accelerated Approval Pathway, Overcoming Resistance to KRAS Inhibition, and More — n-aacr-blog
2026-02-23 — STAT+: Bayer sues J&J over ‘false and misleading claims’ about competing prostate cancer treatments — n-stat
2026-02-23 — STAT+: Pharmalittle: We’re reading about pharma lobbying, a disappointing Novo clinical trial, and much more — n-stat
2026-02-23 — STAT+: Gossamer lung disease drug fails late-stage study, but company will still seek FDA approval — n-stat
2026-02-21 — STAT+: Nature Medicine to investigate study that found cancer treatment is better in morning — n-stat
2026-02-20 — AACR IO 2026 Keynote Highlights: Cancer Vaccines Are Here, and Upgrading T Cells To Thrive in the Tumor Microenvironment — n-aacr-blog
2026-02-20 — STAT+: Element Biosciences launches table-top sequencer capable of $100 genome — n-stat
2026-02-20 — STAT+: Pharmalittle: We’re reading about an FDA official’s speech, a Grail cancer blood test, and much more — n-stat
2026-02-19 — STAT+: Key study of Grail’s cancer detection test fails in setback for company — n-stat
… and 25 more on-topic items.