July 2023 — Monthly Briefing¶
1. What was read¶
763 research articles from 33 of the 33 journals on the roster.
A further 234 records (23% of the 997 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 |
|---|---|---|---|---|
| 460 | 122 | 75 | 328 | 12 |
Volume swing against the median month: cancer -34%, sysbio -50%
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: Nucleic Acids Res 163, Nature 108, JAMA 100, Science 97, Lancet 54, Bioinformatics 48.
2. Net change on the thread curve¶
| Thread | This month | Cumulative | Status |
|---|---|---|---|
cell-therapy |
+12 / −2 | +53 / −17 | forming |
ctdna-mrd |
+2 / −1 | +16 / −10 | splitting |
foundation-models |
+1 / −1 | +15 / −11 | splitting |
immunotherapy-resistance |
+5 / −2 | +71 / −15 | forming |
interception |
+3 | +43 / −4 | forming |
method-credibility |
−3 | +1 / −83 | crisis |
neoantigen-vaccines |
+4 / −1 | +22 / −7 | forming |
neuro-immune-microbiome-axis |
+3 | +44 / −4 | forming |
next-gen-checkpoints |
−4 | +11 / −10 | splitting |
single-cell-3d-genome |
+2 | +13 / −5 | forming |
suppressive-population-inversion |
+5 / −1 | +16 / −15 | splitting contested → splitting |
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¶
A hospital-scale language model and a journal-scale argument about hype, in the
same four weeks. 37286606
(Nature) trains a language model on one health system's own notes and drives
several unrelated clinical predictions off a single backbone — the transfer step
foundation-models rests on, done inside a hospital rather than on a benchmark.
In the same month JAMA runs 37410477
with hype in its title, beside pieces on chatbot privacy
37410449, liability
37410497 and the regulatory gap
37410482, and a letter scoring a
generative model against a diagnostic challenge
37318797. The thread reads +1 / −1
for the month; that is not indecision, it is two literatures publishing past each
other. Worth separating what these are: every AI item in the month is a model
that answers or predicts. Nothing in the month's 763 research articles is an
agent that acts.
Three papers in one journal issue make neurotransmitters the shared currency,
which is what the thread actually claims. Immunity publishes gut lipids
setting peripheral serotonin release through CD1d
37354904, a serotonin metabolite
recruiting eosinophils through GPR35
37279752, and enteric GABA driving
innate gut defence 37437538.
Separately 37380778 (Nature) has
neuronal activity remote from the tumour driving glioma progression through
SEMA4F. neuro-immune-microbiome-axis takes +3 this month, its largest gain
of the quarter — though the quarter is a thin one, and the three months before
it ran +10, +6 and +9. It takes it on the mechanism claim rather than on the
topic, which is the part that counts for a thread whose standing risk is being
a topic.
The atlas layer and the assay layer moved together, and one of them says they are not interchangeable. Nature runs three spatially resolved human organ atlases in one issue — kidney 37468583, intestine 37468586, maternal–fetal interface 37468587 — with cardiac niches 37438528 beside them, and Nature Methods adds gene expression and chromatin accessibility profiled in the same tissue section 37231265. Against that, 37336949 recovers multiway chromatin contacts Hi-C does not return, flagged by its own News & Views 37336950. The output format is consolidating faster than the assays feeding it agree with each other.
Grade rather than novelty: engineered cells reach a randomised comparison
outside CD19. 37272512 (NEJM)
tests a BCMA-directed CAR T product against standard care in myeloma. It is the
strongest single piece of evidence in the month for cell-therapy, and it is the
top of a stack rather than a lone result: the thread closes the month at +12 / −2,
its largest month to date, on CAR-T persistence signatures, c-Kit co-stimulation,
prophylactic anakinra against toxicity, and further readouts in myeloma,
paediatric leukaemia and lymphoma. The two refutations that landed in the same
four weeks (section 4) are constraints on how the platform fails and on how its
early trials report, not a counterweight to that.
Corrected 2026-09-04
This sentence read "the thread still closes the month at −1 net, because two refutations landed on it in the same four weeks", and the point being made was that the month's strongest evidence did not move the curve. The month is +12 / −2. The sign was reversed, and with it the reading: a title-only pass had collected the two refutations, whose titles state their verdicts, and almost none of the twelve supports, whose titles state a mechanism or a trial name. Ranking refutations first while reading titles produces exactly this shape, and it produced it here.
Two detection results land on interception: a
multi-cancer test measured in symptomatic patients referred for investigation
rather than in a screening population
37352875, and microbial metabolites
proposed as markers of an anal precancer
37464040.
4. Refutations¶
(a) Landing on a tracked thread¶
37264229 (Nat Immunol) — how much
the CD94/NKG2-HLA-E axis restrains effector cells is set by an HLA class I signal
peptide polymorphism. The shape of this is what matters: not that a target
failed, but that the size of what blockade could recover belongs to the patient
rather than to the drug. next-gen-checkpoints records nothing of this kind; its
whole ledger is which nominated targets reached a trial.
37392737 (Immunity) — TCR-independent 4-1BB signalling pushes exhausted CD8 T cells into proliferation and terminal differentiation. The sign on an agonist target is inverted.
37011008 (Cancer Discov) — type I interferon signalling through EGR2 drives CAR T cells into intrinsic dysfunction. The failure sits inside the engineered cell, not in the tumour it was aimed at.
37247626 (Lancet) — early-phase CAR T trials against novel target antigens are argued to report activity without a stable denominator. A reporting refutation rather than a biological one, and the only one of its kind this month.
(b) Method credibility¶
Three cards, and the ratio needs saying out loud before section 2's crisis is
misread. This thread is very nearly a ledger of things found unreliable: almost
every card that can land on it refutes, so the cumulative +1 / −83 is mostly the
shape of the instrument and not a measurement of collapse. Read the volume and the
target of the audit, never the sign.
The +1 is worth knowing about rather than rounding off, because a ledger that can only record failures is assuming its answer instead of measuring it. It arrived last month — 37338877 (JAMA), where 91% of 169 nonsignificant trial outcomes turn out to favour the null they were read as failing to establish — and it is one card in eighty-four. That is what a thread with a one-sided intake looks like when it does admit a pass, and it is the only reason the sign can be read at all.
Corrected 2026-09-04
This section read "Two cards ... every card that can land on it refutes by construction, so +0 / −16 is the shape of the instrument". Both halves are wrong. The month is three cards, not two, and the cumulative it quoted as +0 / −16 is +1 / −83 — the title-only pass was seeing about one refutation in five, and the argument built on the undercount read as stable throughout. And the thread is no longer 0-supports: "refutes by construction" was a claim about the card set, and one card in the card set now contradicts it. The instruction that follows — volume and target, never the sign — is what the paragraph was for, and it survives; the absolute it was resting on does not.
36593401 (Nat Biotechnol) — a
genetically encoded MHC-I ligand screen shows that published binding predictors
systematically under-represent cysteine-containing peptides, raising the
predicted ligand count by 9–21%. The bias sits in the prediction step every
neoantigen pipeline runs, which is why the card lands on neoantigen-vaccines as
well.
37349538 (Nat Med, with its News & Views 37400641) — variants published as pathogenic for premature ovarian insufficiency show little penetrance in an unselected population. The gene–disease calls do not survive contact with a denominator.
37224537 (NAR) — disrupting the RNA–RNA contacts a published SPLASH map assigned to influenza genome packaging does not produce the defect that map predicts. A specific published map, tested and not confirmed.
Uncarded but worth the note, because of where in the stack it sits: 37449901 (Bioinformatics) offers a test of whether the populations reported from single-cell RNA-seq are significantly different at all. The audit has reached the basic unit of analysis.
5. Open items¶
Nothing carried forward.
6. Meetings¶
1 abstracts from 1 meeting.
| Meeting | Abstracts | With full text |
|---|---|---|
| ASCO Annual Meeting 2023 | 1 | 1 |
7. Regulatory¶
| Event | Count |
|---|---|
| trial-first-posted | 433 (of which 380 genuinely oncology) |
| ema-authorisation | 3 |
| fda-approval | 2 |
2 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:
- 2023-07-20 — FDA approves quizartinib for newly diagnosed acute myeloid leukemia
- 2023-07-31 — FDA approves dostarlimab-gxly with chemotherapy for endometrial cancer
2 EMA oncology authorisations:
- 2023-07-04 — Lytgobi
- 2023-07-07 — Columvi
8. Trade press¶
No items. The press archive covers 2024-01 to 2026-08; feeds that cannot be paginated backwards contribute only from the first harvest forward.