January 2023 — Monthly Briefing¶
1. What was read¶
991 research articles from 33 of the 33 journals on the roster.
A further 186 records (16% of the 1,177 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 |
|---|---|---|---|---|
| 391 | 176 | 74 | 518 | 18 |
Volume swing against the median month: bioinfo +91%, sysbio -25%
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 250, Bioinformatics 174, Nature 99, Science 77, JAMA 77, J Clin Oncol 61.
2. Net change on the thread curve¶
| Thread | This month | Cumulative | Status |
|---|---|---|---|
cell-therapy |
+6 / −3 | +6 / −3 | forming (opens as forming) |
ctdna-mrd |
+3 / −4 | +3 / −4 | contested (opens as contested) |
foundation-models |
+1 / −2 | +1 / −2 | too-early (opens as too-early) |
immunotherapy-resistance |
+6 / −2 | +6 / −2 | forming (opens as forming) |
interception |
+6 | +6 / −0 | forming (opens as forming) |
method-credibility |
−9 | +0 / −9 | crisis (opens as crisis) |
neoantigen-vaccines |
+3 / −1 | +3 / −1 | too-early (opens as too-early) |
neuro-immune-microbiome-axis |
+5 / −1 | +5 / −1 | forming (opens as forming) |
next-gen-checkpoints |
+1 / −3 | +1 / −3 | too-early (opens as too-early) |
single-cell-3d-genome |
+2 / −2 | +2 / −2 | too-early (opens as too-early) |
suppressive-population-inversion |
−4 | +0 / −4 | too-early (opens as too-early) |
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¶
Written in arrears, and re-read from abstracts
2023 was added to the corpus after 2024, 2025 and 2026 had already been read, which is the opposite of the position the monthly practice is supposed to simulate. That has not changed and still caps what these months can support. What has changed is the reading. This month was first carded from titles and journal names alone, then re-scanned in full from abstracts on 2026-09-04 with the instrument 2026 was read with. 2023 went from 25.1 to 80.4 cards per thousand research papers, against 2026's 84.8; 40% of this month's cards state something no title carries. The earlier version of this note said the thin record was a floor under the threads. It was a property of the instrument, and roughly three cards in four here were invisible to it.
Plasma sequencing reads out resistance three times in one journal in one month. 36007218 (JCO) resolves the acquired-resistance landscape to anti-EGFR therapy in metastatic colorectal cancer; 36067452 does the same on the cetuximab arm of CALGB/SWOG-80405; 36351210 reports that the mechanisms differ by regimen and by line of therapy. Three independent groups, one target, one month is the strongest evidence grade literature-only surveillance can produce, and what they converge on is ctDNA answering what next rather than how bad. 36646802 (Nat Med) ties molecular residual disease to whether adjuvant chemotherapy works at all in colorectal cancer.
Engineered cells reach patients in two compartments the platform is not built for. 36259971 (Cancer Discov) delivers a B7-H3 CAR intraventricularly into diffuse intrinsic pontine glioma and reports bioactivity; 36624315 (Nat Med) takes an HLA-restricted TCR against a cancer-testis antigen through a phase 1 across solid histologies. Both are small and single-arm, and neither is an efficacy claim. What they establish is that the delivery route and the manufacturing exist off the B-cell surface.
Both non-lymphoid arms of the neuro-immune-microbiome thread run at top tier in the same month. 36517594 (Nature) reports glioma networks with autonomous rhythmic electrical activity driving their own growth — neural excitability as a tumour-intrinsic growth variable rather than as context. 36563680 (Cancer Cell) puts gut bacterial nucleotide synthesis on the causal path of neoadjuvant chemoradiotherapy failure in rectal cancer. The thread's claim is that these are one mechanism network and not three adjacent fields; this month they are at least at one publication tier.
A premalignant state with a mechanism attached. 36656933 (Science): a
BTG1 mutation makes B cells supercompetitive and primes them for transformation.
36098678 (Cancer Discov) reports cell competition shaping metastatic
latency and relapse in the same month. interception needs states that are
measurable and mechanistically distinct from cancer, and clonal competition is
the form both of these take.
Foundation models open on a resource and two failures. The only support is a
resource announcement — 36420884 (NAR), the RCSB PDB now serving a million
machine-computed structures alongside solved ones. Model output entering the
reference layer that everything downstream reads is a real change in the
substrate, and it is not a result. The two findings the month does carry both run
against the models: 36355460
(Bioinformatics) optimises multiple sequence alignments end-to-end against
AlphaFold's own confidence and gets self-inconsistent, adversarial inputs, so the
confidence score can be gamed; 36692561
(JAMA) finds machine learning adds nothing over the pooled cohort equations for
stroke risk and performs worse in Black individuals. foundation-models opens
+1 / −2: a resource on the supporting side, and results only where the models
fail.
Corrected 2026-09-04
This paragraph said the month's entire signal on the thread was the PDB resource plus news coverage of a chatbot, and that nothing in the 1,177 titles read this month was a foundation-model finding. Two of them are, and the 2026-09-04 abstract re-scan carded both. Neither title states its verdict — one reads as a method paper, the other as an accuracy comparison — which is how a title-only pass came to read the month as empty rather than as adverse. The reading to withdraw is "thin", not the point about the resource.
4. Refutations¶
(a) Landing on a tracked thread¶
Three of this month's cards invert a population the field is organised around
removing, which is why suppressive-population-inversion opens at +0 / −4 — the
fourth card being 36577929 (Nat
Immunol), where deleting PD-1 destabilises the lineage identity of
tumour-infiltrating regulatory T cells rather than only releasing the CD8
compartment. next-gen-checkpoints opens +1 / −3 on much of the same material;
its single support is 36585453 (Nat
Cancer), where blocking 4-1BB, LAG-3 and CXCR1/2 together produces durable
responses in pancreatic cancer that PD-1 blockade alone does not.
Corrected 2026-09-04
This sentence said both threads open at zero support. next-gen-checkpoints
opens +1 / −3: the abstract re-scan carded a combination-blockade result that
a title-only pass had not collected. suppressive-population-inversion does
open with no support, on four refutations rather than the three described
below.
36652534 (Sci Transl Med): CD200-positive cytotoxic T lymphocytes are required for anti-PD-1/PD-L1 therapy to work. CD200 is an inhibitory ligand and its axis is a standing checkpoint nomination; here the cells carrying it are the effector, so blocking it removes what the therapy needs.
36525971 (Cancer Cell): IL-5-producing CD4 T cells and eosinophils cooperate to improve checkpoint response in breast cancer. Type 2 immunity is routinely nominated as tumour-promoting; this is that nomination running the other way, in a randomisable direction.
36574773 (Immunity): CD39, the adenosine-axis ectonucleotidase pursued as a blockade target, identifies the tumour-reactive CD8 compartment and predicts checkpoint response in human lung cancer. In the same month 36040496 (Blood) reports CD39/CD73 dysregulation driving T-cell immunosuppression in Sézary syndrome. The molecule is being read as an effector marker and as a suppressive effector at once, and this is only the opening half of that disagreement.
Three further results strain their threads from inside. 36638191
(Sci Immunol) shows that whether an anti-PD-1 antibody antagonises or agonises
the receptor is set by which epitope it binds — a variable a target-picking
programme does not control. 36646856 (Nat Cancer) locates checkpoint
resistance in an interferon-associated epigenetic memory the cancer cell carries
itself, which is the tumour-intrinsic explanation immunotherapy-resistance
claims the field is moving away from. And 36070539 (JCO) is the month's
cancer vaccine with an efficacy readout: a universal shared-antigen peptide
vaccine in refractory NSCLC, running on exactly the antigen space the neoantigen
claim says is insufficient.
The one card pushing the other way is 36563676 (Immunity), which puts checkpoint resistance in the humoral compartment: the therapy itself alters IgG glycosylation and the sialylated antibody blunts type I interferon.
(b) Method credibility¶
Nine cards this month. The three described here escalate through different objects, and none of those three is a model.
36522432 (Nat Genet): guide RNAs validated against the reference genome behave differently in different people, so a therapeutic editor's off-target profile does not transfer from the reference to the patient. A reagent failing at the level the clinic will meet it.
36635510 (Nature): scientists cannot reliably distinguish machine-generated abstracts from real ones. What is being audited is the readable surface of the literature — the thing every downstream curation step, this corpus included, treats as ground truth. 36653617 and 36624237 run beside it.
36625810 (JAMA): devices cleared by claiming equivalence to earlier devices that had themselves been recalled are more likely to be recalled in turn, with 36625811 on which submission characteristics predict it. The object here is neither a model nor an assay but the regulatory chain that certifies instruments.
method-credibility reads −9 and will keep reading negative. Its denominator
field in threads.yml says why: the thread is a ledger of things found
unreliable, so the sign is very nearly fixed before any reading happens — 93% of
its cards across the whole corpus refute — and the ratio carries almost no
information. Read the volume and the object, never the sign.
Corrected 2026-09-04
Two figures here were wrong. The month reads −9, not −3. The five cards the
original pass missed are wild-type Cas13's collateral transcriptome cleavage,
a spatial-transcriptomics deconvolution benchmark, a fresh-versus-frozen
protocol artefact in single-cell composition, and the two model audits
described in section 3 — so the month's objects do include models, even
though the three described above do not. And "every card on it refutes by
construction" was already untrue when it was written: threads.yml now
records 25 supports among 654 cards, and a ledger
that cannot record a pass is assuming its answer rather than measuring it.
The instruction — read the volume and the object, never the sign — is
unchanged.
5. Open items¶
Nothing carried forward.
6. Meetings¶
No abstract book published this month. 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 | 410 (of which 354 genuinely oncology) |
| fda-approval | 5 |
| ema-authorisation | 3 |
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:
- 2023-01-19 — FDA approves zanubrutinib for chronic lymphocytic leukemia or small lymphocytic lymphoma
- 2023-01-19 — FDA grants accelerated approval to tucatinib with trastuzumab for colorectal cancer
- 2023-01-26 — FDA approves pembrolizumab as adjuvant treatment for non-small cell lung cancer
- 2023-01-27 — FDA approves elacestrant for ER-positive, HER2-negative, ESR1-mutated advanced or metastatic breast cancer
- 2023-01-27 — FDA grants accelerated approval to pirtobrutinib for relapsed or refractory mantle cell lymphoma
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.