December 2023 — Monthly Briefing¶
1. What was read¶
654 research articles from 33 of the 33 journals on the roster.
A further 235 records (26% of the 889 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 |
|---|---|---|---|---|
| 415 | 158 | 75 | 224 | 17 |
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: Science 92, JAMA 88, Nature 62, Bioinformatics 51, Nat Med 44, Nucleic Acids Res 42.
2. Net change on the thread curve¶
| Thread | This month | Cumulative | Status |
|---|---|---|---|
agentic-ai |
+2 | +2 / −0 | too-early (opens as too-early) |
cell-therapy |
+7 / −2 | +85 / −25 | forming |
ctdna-mrd |
+2 | +24 / −12 | forming splitting → forming |
foundation-models |
+4 / −5 | +32 / −20 | splitting |
immunotherapy-resistance |
+5 / −2 | +98 / −31 | forming |
interception |
+6 / −2 | +60 / −9 | forming |
method-credibility |
−12 | +2 / −129 | crisis |
neoantigen-vaccines |
+2 / −1 | +37 / −15 | forming |
neuro-immune-microbiome-axis |
+2 / −2 | +60 / −7 | forming |
next-gen-checkpoints |
+1 / −3 | +24 / −17 | splitting |
single-cell-3d-genome |
+4 / −1 | +25 / −7 | forming |
suppressive-population-inversion |
−2 | +23 / −28 | 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¶
Nature reports autonomous chemical research driven by large language models (38123806), and Science reports an autonomous molecular-discovery loop three weeks later (38127734). These are different claims from a model that answers questions well. In the Nature system the model plans, executes and interprets experiments through external tools and instruments; the Science system closes the loop from prediction to physical measurement and back without a human deciding each iteration. Nature also carries an autonomous synthesis laboratory in the same month (38030721) with no language model in it at all, which is a useful control: the robotics half and the reasoning half are being built by different communities and are separable. agentic-ai reaches two stanced cards this month and the status rule calls it too-early, correctly — two papers is a pair, not a thread, and nothing here shows any of these systems running as infrastructure that other laboratories depend on.
Corrected 2026-09-04
This read "agentic-ai opens this month at two cards". It does not open
here. The abstract re-scan found the thread's first card in 2023-03 — a
NEJM appraisal of GPT-4 as a clinical chatbot, carded neutral — and
threads.yml now records opened: 2023-03.
The generated table in section 2 still prints (opens as too-early) beside
this month, and that is not a contradiction: the derived marker fires on the
first card carrying a stance the rule counts, and status_of ignores
neutral entirely. Two senses of "opens" — first card, and first card the
verdict can see — and only one of them recomputes. Worth knowing before
reading any (opens as …) marker in this corpus as a date.
Scaling produces one unambiguous win, in a domain with physical ground truth. Nature reports a large-scale deep learning campaign expanding a materials discovery space by orders of magnitude (38030720). Read this against section 4(b) below, which is the same month's evidence that the same strategy fails on human genetic variation; the honest reading of December is that scale delivers where the target is a physical structure and struggles where the target is a person. Nature Medicine's population-scale detection of pancreatic cancer on scans not acquired to look for it (37985692) is the clinical version of the win, and lands on interception as much as on foundation-models.
The host compartment supplies three separate handles on checkpoint response. Nature Medicine associates a patient's pre-treatment emotional distress with response to neoadjuvant checkpoint blockade in melanoma (37957378) — a neuroendocrine host variable entering the resistance account at the clinical level, not the mouse level. JITC reports early C-reactive protein kinetics predicting response and mortality across cancer types (38097343), extending October's acute-phase result to a cancer-agnostic setting. And Nature Immunology has tumour-released kynurenine biasing progenitor differentiation toward megakaryocytes in patients (37919525), which is the tumour reprogramming host haematopoiesis at a distance. JITC separately reverses resistance by targeting an integrin on fibroblasts rather than anything on the tumour (38040421).
Neurotransmitters keep turning up as immune currency. Immunity shows dopamine-receptor signalling on gut epithelium setting the severity of central nervous system autoimmunity (37992711) — a transmitter receptor in the gut controlling disease in the brain, which is the gut–brain axis stated as a mechanism rather than as a metaphor. Nature Immunology reports the interferon response depleting serotonin (37996765), immune signalling consuming a transmitter rather than merely responding to one. JCO also publishes a randomised, double-blind, placebo-controlled trial of faecal microbiota transplantation in allogeneic transplant and AML (37235836); its title states no direction, so it is carded neutral, but a placebo-controlled clinical test of this thread existing at all is worth recording, because the thread is often described as mechanistically dense and clinically untested.
Precursor states are being read genomically, and residual disease is deciding therapy. Cancer Cell runs two papers in one issue on the precancerous window — genomic alterations driving precancerous to cancerous lesions in oesophageal cancer (38039962) and clonal dynamics of gastric progression through intestinal metaplasia (37890493) — treating the precursor as a staged object rather than a category. Science Translational Medicine gets high-grade serous ovarian cancer out of the DNA in a Pap test (38055801). On the residual-disease side, Blood uses lineage-specific measurement to predict relapse in patients stopping therapy for chronic myeloid leukaemia (37616555), which is a molecular signal deciding whether treatment continues, and Annals of Oncology reports a plasma assay needing no tumour tissue at all (37879442).
4. Refutations¶
(a) Landing on a tracked thread¶
The depletion strategy is defeated by the response it provokes. Science Immunology reports that after regulatory T cell depletion, conventional T cells acquire suppressive function, and that this limits antitumour immunity (38100544). This is the cleanest refutation in the quarter and deserves to be read carefully, because it is not the usual complaint. The usual complaint about suppressive-population-inversion is that depleting a suppressive population has off-target costs elsewhere — autoimmunity, impaired repair — which is a toxicity argument and can in principle be engineered around. This is a different and worse objection: the suppression is a function the system restores, not a property of the cells that were removed. Take the cells away and the role is reassigned. If that generalises, then the entire class of interventions built on identifying a suppressive population and deleting it is aiming at the wrong level of description, and no amount of better targeting fixes it. The thread's claim is that depleting or blocking such a population improves antitumour outcomes; this says the manoeuvre can leave the tumour exactly where it was. It is a single paper in mouse models and should not be over-read — but it is the kind of result that, if it holds, invalidates a strategy rather than a target.
CD47 gets nominated and qualified in the same month by different journals. Nature Immunology reports CD47 masking pro-phagocytic ligands in cis to suppress antitumour immunity (37945822) — a mechanistic case for blocking it. Cancer Cell reports that the antitumour activity of anti-CD47 antibodies requires Fc–FcγR engagement (37977147) — so removing the inhibitory signal is not sufficient, and the blockade rationale as usually stated is incomplete.
JITC shows an intracellular checkpoint trading one limit for another: deleting the phosphatase PTPN22 in T cells for adoptive therapy improves effector function and promotes exhaustion in the same cells (38056892). The Lancet reports that the regulator has opened a safety investigation into recipients of CAR T-cell therapy (38071978), which converts scattered case reports into a class-wide question and is a fact about the platform rather than about any one product. Nature Genetics asks directly whether the best-characterised clonal precursor state is consequential at all (37996619) — interception assumes precursors matter enough to treat, and that assumption is being questioned in the field's own journal. And Nature Medicine argues clinical models must convey per-patient predictive uncertainty (37821686), naming a property current systems do not have.
(b) Method credibility¶
Nature Genetics publishes three titles in one month saying that genomic deep learning fails on variation between individuals. Personal transcriptome variation is poorly explained by current models (38036790); an independent benchmark of deep networks predicting personal gene expression from sequence reports shortcomings (38036778); and a synthesis states that current approaches struggle to capture human genetic variation as a class (38036789). This is the strongest thing in December and it belongs in this section rather than section 3, because its object is a whole model class rather than any thread's biology — though it lands squarely on foundation-models too, and is three of that thread's five refutations this month.
What makes it serious is the axis of failure. These models are trained and benchmarked on variation across the genome and are evaluated by how well they predict expression at a held-out locus; the quantity that matters for a person is variation between individuals at the same locus, which is precisely where they fall over. A benchmark can be excellent and still measure the wrong thing, and one journal reaching that conclusion three times in one month through independent work is a different quality of evidence from a single negative paper.
Nature Cancer finds that most approved combination therapies for advanced cancer are explained by simple additivity (37974028) — the synergy routinely claimed for these regimens is largely absent when tested against a null, which is a statement about a very large published literature and about how combinations are justified. Genome Research revises the structural model of sperm chromatin twice in one issue: cohesin and CTCF do not assemble TADs in Xenopus sperm and male pronuclei (38129077), alongside a reconsideration of chromatin packaging in mouse sperm (38129076). And the Journal of Experimental Medicine issues an Expression of Concern on an immuno-oncology mechanism paper (38015790) — one concrete instance behind this thread rather than a general worry about reproducibility.
How to read the ratio¶
Twenty-eight of this month's sixty-four cards refute, and the quarter runs near forty per cent — above the corpus's established rate. That is a property of how these months were read rather than of the months: the reading brief ranked refutations first and went looking for them, and a reader comparing December 2023 against other months in this corpus is partly comparing two reading practices. Two further limits belong on the page. These months were read against titles alone at first, so a stance was recorded only where the title stated one — the Nature Genetics trio above is carded precisely because all three titles say what they found. The quieter papers that sentence said were missing were gone looking for in an abstract re-scan on 2026-09-04; 44% of this month's cards came back stating something no title carries, and the month went from twenty-four cards to sixty-four. The refutation share barely moved across that — eleven of twenty-four, then twenty-eight of sixty-four — so what the title-only reading got wrong here was the size of December, not its temper. And they were read in arrears. Every card is written to what the paper itself supports in December 2023; that is a discipline applied by hand, not one the method enforces.
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 | 337 (of which 309 genuinely oncology) |
| ema-authorisation | 8 |
| fda-approval | 4 |
4 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:
- 2023-12-01 — FDA grants accelerated approval to pirtobrutinib for chronic lymphocytic leukemia and small lymphocytic lymphoma
- 2023-12-13 — FDA approves eflornithine for adult and pediatric patients with high-risk neuroblastoma
- 2023-12-14 — FDA approves belzutifan for advanced renal cell carcinoma
- 2023-12-15 — FDA approves enfortumab vedotin-ejfv with pembrolizumab for locally advanced or metastatic urothelial cancer
1 EMA oncology authorisation:
- 2023-12-07 — Elrexfio
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.