104 open register · 0 complete · 85 partly written · 19 not yet written
Open register
What is not understood: things observed and unexplained, things expected and never seen, Amit's own questions, and questions readers asked on a page. Repeat questions are answered first. A question that closes moves to the solved table.
-
A trial enrolment figure that disagrees inside one sponsor's own materialTwo numbers for how many people were enrolled in the tominersen GENERATION HD1 trial appear in material from the same sponsor. big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- Which figure is the enrolment and which is something else (randomised, treated, analysed)? Check the registry entry and the published paper, not a press page.
- Does any reported result change depending on which denominator is used?
-
Are histone marks mostly a cause of gene regulation, or a record of it?Whether histone modifications drive differences between chromatin states or mainly result from transcription and nucleosome remodelling is argued both ways, with causal evidence at few loci. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. One camp argues marks mostly follow transcription and remodelling. A review says some are causal and others record. The best causal example is one Drosophila Hox gene.
-
Are orthology and evolutionary relationships first-class in biological ontologies?Yes in four resources (RO, ORTH, Biolink, the Alliance); in GO annotation it is a method, and the vocabulary is not joined to the confidence. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Are protein complexes best thought of as "megaproteins"? amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- How do complexes form and know when to form? Does treating a complex as one combinatorial protein help or mislead?
-
Can a biomedical knowledge-graph edge say what happens under intervention?Every edge read in public graphs is an association or a literature assertion; an interventional edge exists only where perturbation data exist, and no graph read tests that. big-open open-question not yet written not yet written not yet written not yet written partly written
-
Can a gene be silenced permanently, and how does cellular memory work?Can a gene be silenced permanently? In opened sources silencing lasts as long as machinery re-makes it, the longest measured engineered case is nearly a year in mouse liver, and every loop found has been undone by something. amit observed-unexplained not yet written not yet written not yet written not yet written partly written
Short answer from the sources. Silencing can last through many divisions and, engineered in mouse liver, for nearly a year. In each tested case it is kept by running machinery and something could undo it. No opened source shows a state that cannot be undone, or shows that none can exist.
His hypothesis (amits-hypothesis): can a gene be silenced for ever. In his words: "there are a lot of phenomena that we still don't fully understand, like maybe cellular memory. How does that work? We know that there's gene silencing caused by permanent epigenetic regulations. In theory, could you permanently silence a gene? In theory, it makes sense, right?"
His hypothesis (amits-hypothesis): the knife picture. In his words: "If you think about a person with two arms and a knife, and they cut off both of their arms using the knife, now they can't cut off anything else. They've cut off the ability to cut off anything else, right? You can imagine this as a forever impact."
-
Can a gene delivered by AAV be turned down or off after dosing in people?Regulated expression is well described as a design goal, but no opened source shows a clinical AAV product with a working off-switch. big-open open-question not yet written not yet written needs-research not yet written partly written
The question in brief. A promoter sets how strongly a delivered gene is read; how long it lasts comes from the form of the vector DNA.
-
Can a histone mark carry memory through cell divisions without a DNA sequence to rewrite it?Two Drosophila studies on Polycomb repression report that H3K27me3 nucleosomes carry a memory of repression through replication, and differ on whether new nucleosomes can be marked without a sequence anchor. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. Two Drosophila studies agree that marked parental nucleosomes pass to daughters and can keep a gene repressed for a few replications. They differ in emphasis on whether new nucleosomes get marked without a sequence anchor. No mammalian test was found.
-
Can a pocket or an interface be defined as a typed entity in logic, or are they ensembles?A pocket was not found as a typed entity anywhere read; an interface exists as a typed record inside one database (MDposit) but not as a vocabulary or ontology term. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Can an agent fetch the variant answer key from a public database during a run?No benchmark opened blocks lookup of public variant databases, and no measurement of agents doing it was found. big-open open-question not yet written not yet written not yet written not yet written partly written
What is known and what is not?. One benchmark gives no tools and one does not state its network access in the text read. Whether agents fetch public answer keys has not been measured in sources opened.
-
Can paired host-genetics and pathogen-genomics results share one schema with interaction resources?Paired studies exist for five pathogens; interaction resources overlap little and drop about a quarter of proteins on identifier mapping; no shared schema joins them. amit open-question not yet written not yet written needs-research not yet written partly written
-
Can parts of the genome be traced to specific events such as an ice age? amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- Which popular examples of traceable adaptation hold up, and which do not? Give the evidence for and against each.
-
Can the reason for asking be stored as structure so it changes rankings beyond re-ranking hubs?Purpose lives in tool defaults, such as the Open Targets safety section, not as a stored edge, and no benchmark tests a purpose-conditioned ranking against a purpose-blind one. amit open-question not yet written not yet written needs-research not yet written partly written
-
Could one model cover every step of drug discovery?Whether one model could cover every step of drug discovery; the sourced material here bears on baselines and on how little a single readout predicts. amit open-question not yet written not yet written not yet written not yet written partly written
-
Do epigenetic clocks measure biology or assay noise?Replicate noise moves a clock reading by up to 9 years, and published clocks disagree on reprogrammed cells, so the size and even the direction of a reported rejuvenation depend on the clock. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. Replicate noise alone shifts six clocks by up to 9 years. On reprogrammed cells, published clocks gave about 40 years younger to about 13 years older. The size of any rejuvenation depends on the clock; none is shown to be wrong.
-
Do KinCore, KLIFS and other kinase state schemes agree where they overlap?No agreement study was found; the KinCore authors say the criteria are not widely agreed. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Do people make antibodies to PEG on carriers, and do they clear the carrier faster?Anti-PEG antibody effects on lipid nanoparticle clearance and reactions were seen only in a search summary in report 5 and are not held. big-open open-question not yet written not yet written not yet written not yet written partly written
-
Does a quantum-derived representation add anything to an ECFP fingerprint on a biological endpoint?Quantum-informed atom pairs gained 0.5 to 6% over plain atom pairs, not over ECFP; no scaffold-split head-to-head against ECFP on an activity task was found. amit open-question not yet written not yet written needs-research not yet written partly written
-
Does an AI judge agree with scientists when scoring a biology agent's lab notebook?No measured agreement between an AI judge and a scientist on biology lab notebooks was found. big-open open-question not yet written not yet written not yet written not yet written partly written
What is known and what is not?. Judge agreement is measured for chat and for idea ranking, and the best result on expert-level ranking is near chance. For biology lab notebooks no measurement was found.
-
Does any RNA plus protein integration method beat RNA-only and concatenated-PCA baselines in an independent test?No independent head-to-head of WNN, MOFA+ and totalVI with RNA-only and concatenated baselines on RNA plus protein, with labels not derived from them, was found. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Does assay metadata beyond a batch label improve prediction on held-out labs?Batch labels are used and often subtracted; operator, site, date or quality as typed inputs have no measured gain on record. amit open-question not yet written not yet written needs-research not yet written partly written
-
Does changing genome folding alone change gene expression?Removing cohesin erases loops yet changes only a minority of active genes, and folding changes are hard to separate from other effects, so cause and effect between folding and expression remains open. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. Loops vanish when cohesin is lost, yet compartments and marks stay and only a minority of active genes change. Contact is not regulation. No source in the set isolates folding alone at a gene.
-
Does changing only the data representation change performance, with data and algorithm held fixed?A test of the claim that algorithms are limited by their representation and representation by their data: this atlas found no clean representation-only comparison. amit open-question not yet written not yet written needs-research not yet written partly written
His claim. "You have innovation at the algorithm layer. I find you can innovate at the algorithm layer all you want, but that can only be as powerful as the data representation. The data representation, of course, can only be as powerful as the data, but at least there's more space to innovate there"
-
Does chemistry pretraining still help enzyme-reaction prediction when the test enzymes are unlike the training enzymes?Whether the measured gain from patent-chemistry pretraining survives a split by enzyme family or sequence cluster is untested. big-open open-question not yet written not yet written needs-research not yet written partly written
What do the measured gains show?. Pretraining on patent chemistry measurably helped on random splits: about 10% patent-only, about 20% enzyme-only and about 50% mixed in one study. Whether it helps on enzymes unlike the training set was not found.
-
Does choosing a representation per query beat one fixed representation?Per-task winners and concatenations are documented; a trained per-query chooser compared with a tuned fixed fingerprint on a scaffold split was not found. amit open-question not yet written not yet written needs-research not yet written partly written
-
Does epigenetic inheritance across generations occur in mammals?Whether epigenetic marks, not DNA sequence, carry an environmental effect from parent to offspring in mammals is read as demonstrated by one set of authors and as unproven by others. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. Mouse germ cells erase most methylation, and a few elements escape. One review sees inheritance in many species with a smaller mammalian effect; another source finds the mammalian evidence inconclusive. They agree a mechanism could exist and differ on whether an effect is shown.
-
Does in vitro screening kill projects that would have worked?No opened source measures false-negative kills by in vitro assays. amit open-question not yet written not yet written not yet written not yet written partly written
-
Does LLM-judged hypothesis ranking beat a trivial baseline, and has any builder reported one?Does LLM-judged hypothesis ranking beat a trivial baseline, and has any builder reported one? big-open open-question not yet written not yet written not yet written not yet written partly written
What is known. In an independent study the best LLM pairwise ranker agreed with human scores 53.3% of the time, humans with each other 56.1%, random 50%. Sakana reported random and always-reject baselines for its reviewer. The co-scientist's Elo is not based on independent ground truth in its authors' words.
-
Does matching therapy to a child's leukemia genome improve outcomes?Whether treating relapsed pediatric leukemia with the drug that matches its alteration helps children, which no study read has tested against unmatched care. big-open open-question not yet written not yet written not yet written not yet written partly written
What the figures mean. The 77% in the registry, the 18% of recommendations at the two highest evidence tiers in the paper, and the 17 treated patients answer different questions. None of them is a response rate.
-
Does one unhealthy cell change the cells around it? amit observed-unexplained not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- Do cells share DNA or release signals that change the regulation of neighbours? Find the studies he half-remembers and say what they showed.
-
Does poly(A) tail length set translation, and does the closed loop exist everywhere?A longer tail gives more protein in early embryos but not in most cells, and the closed-loop model that joins cap and tail is partly unresolved. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Does repeated exposure raise the enzymes that clear a drug (induction)?Whether repeated dosing raises liver or gut enzyme levels, by how much and how fast, was not covered by any source opened in report 5. big-open open-question not yet written not yet written not yet written not yet written partly written
-
For how many human genes does a second start codon make a second protein with its own job?For how many human genes does a second start codon make a second protein with its own job? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. Named cases exist (CEBPA, PTEN), thousands of footprint-called starts exist, and in Jurkat cells N-terminal proteomics confirmed alternative starts for 133 genes. No published share of human proteins with a confirmed second N-terminus was found.
-
Forced selectivity: a drug that flags itself when it binds the wrong target amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- Has anything like this been built? What host machinery could read such a flag inside a cell, and what is the closest real mechanism?
-
Has a measured protein interaction ever been shown false and stopped a named drug programme?Has a measured protein interaction ever been shown false and stopped a named drug programme? amit open-question not yet written not yet written not yet written not yet written partly written
-
Has an ontology-based diagnosis been shown to change treatment, survival or cost?Documented evidence is rank in solved cases and diagnostic yield; no outcome study was found. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Has any top prediction from a public biomedical graph been tested prospectively?No prospective test of a top prediction from Hetionet, PrimeKG, SPOKE or the Translator graphs was found. big-open open-question not yet written not yet written not yet written not yet written partly written
-
Has anyone measured kon and koff for interactions at proteome scale?Has anyone measured kon and koff for interactions at proteome scale? amit open-question not yet written not yet written not yet written not yet written partly written
-
Has compliance with any omics data standard (GA4GH, HUPO-PSI, Human Cell Atlas, HuBMAP) been measured?No compliance audit of GA4GH, HUPO-PSI, Human Cell Atlas or HuBMAP specifications was found; related audits cover microarray repeatability and BioSample metadata. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Has every protein been tested against every other protein? amit open-question not yet written not yet written not yet written not yet written partly written
Short answer. No. The largest binary map, HuRI, reports 52,569 interactions among 8,275 proteins, out of a space of over 150 million pairs, and its authors put it at 2 to 11 percent of the binary interactome. The largest co-complex map, BioPlex 3.0, used 10,128 baits in one cell line. Nobody claims completeness.
-
How are a dose and a schedule chosen from the PK/PD models?A first-in-human dose is an animal-derived safety margin; the label dose comes from escalation and exposure-response work, and oncology doses have often needed later correction. big-open open-question not yet written not yet written answered not yet written partly written
The short answer. A first-in-human dose is a safety margin, not a dose predicted to work. FDA 2005 takes the highest animal dose with no adverse effect, scales it by body surface area and divides by 10; EMA 2017 takes the lowest of several estimates, including the dose expected to give the first biological effect. In patients doses are raised by a rule until toxicity limits them, and later work picks the label dose from exposure and response.
-
How could loneliness or emotional harm damage cells? amit observed-unexplained not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- What do the studies he mentions actually show, how strong is the evidence, and is any molecular route established?
-
How do we know a pathway is a bona fide pathway? amit open-question not yet written not yet written not yet written not yet written partly written
Short answer. A curated pathway is a chain of reactions, each tied to a paper that showed that step in humans by direct experiment, reviewed by a second curator and outside experts. A screen-derived interaction is not a reaction. A large share of what a pathway database stores is inferred, not observed.
-
How do you detect wrong merges between same-category nodes in a knowledge graph?RTX-KG2 flags merged clusters with clashing categories, and its authors say the check misses wrong merges between compatible categories. big-open open-question not yet written not yet written not yet written not yet written partly written
-
How faithfully is DNA methylation copied at each division, and does it depend on the CpG context?How faithfully a CpG site keeps its methylation through one cell division, and whether isolated sites, clustered sites, islands and non-islands differ, has no published per-site, per-division number. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. One locus gives about 0.96 per site per division, refit to a failure of 0.024 once conversion error is modelled. No source splits the rate by CpG context. Two methods point the same way: isolated CpGs are copied worst.
-
How far is a typical human cell from the nearest capillary, measured?Measured human distances are tens of micrometres in muscle and about 76 to 105 um cell to vessel in oral cancer; human cortex has only a model and liver no measured value. big-open open-question not yet written not yet written needs-research not yet written partly written
The short answer. No single number answers this, and only some distances were measured. In human thigh muscle each capillary supplies a territory of about 3,233 square micrometres. In two human tumour series the distance was 76 to 105 um from cell to vessel and 304 um from vessel to vessel. For human cortex only a computer model survives checking: 73% of tissue within 40 um of a vessel. For liver there is no measured value.
-
How long do X inactivation and imprints last, and how often do they fail?No opened primary study measures how long an inactive X or an imprint lasts in years or per cell division; what exists is snapshots of leakage, perturbation rates and cultured stem-cell erosion. big-open open-question not yet written not yet written not yet written not yet written partly written
-
How long does a human RNA polymerase II wait at the pause?How long does a human RNA polymerase II wait at the pause? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. Published estimates for the same step range from about 42 seconds (live-cell imaging, MRC-5 fibroblasts) to about 7 minutes (HEK293 run-on) and over 15 minutes at some Drosophila genes; a review gives about 2 to 30 minutes and calls it a controversy.
-
How many fewer function and mechanism annotations are there than pathogenicity annotations? amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- Verify the real ratios between variant pathogenicity annotations, functional annotations and molecular mechanism annotations, with sources. His tenfold steps are a guess.
-
How many genes in the simplest cell do we still not understand, and what does that mean?How many genes in the simplest cell do we still not understand, and does that signal a missing principle or an annotation backlog? amit open-question not yet written not yet written not yet written not yet written partly written
His claim. He said: "there are actually still around 70 to 90 genes that we still don't fully understand, right? That, to me, is a sign that there's some deep biological issue that we can't fully explain."
What the sources say about his number. 70 to 90 matches one count of one cell: the syn3A 'unclear function' genes, 91 in 2019 and 87 in 2022. No source defines a set of 70 to 90. Other schemes give other counts, so a count is a statement about a labelling rule as much as about biology.
-
How many human protein-coding genes still have no real name?HGNC counts 427 protein-coding genes on placeholder symbols; that is not the same as how many are unknown. amit open-question not yet written not yet written not yet written not yet written partly written
His thought. He said there may be very rare genes that are not yet fully understood or have not even been named.
-
How many published enrichment results change under a correct background and a current release?The instability of enrichment with the GO release is measured; how many published conclusions it overturns is not. big-open open-question not yet written not yet written needs-research not yet written partly written
-
How many spontaneous and damage reactions are missing from standard metabolic reconstructions?Practitioners say damage and repair chemistry is missing from reconstructions, but no count of what is missing was found. big-open open-question not yet written not yet written needs-research not yet written partly written
What has been measured so far?. One review reports a 41% match between observed and predicted metabolites in infected red blood cells, and a rule set that predicts 180,891 spontaneous reactions, a search-space size and not an accuracy.
-
How many translated noncanonical ORFs make functional products?How many translated noncanonical ORFs make functional products? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. Of 7,264 human Ribo-seq ORFs in one catalogue, 609 had mass-spectrometry support and 75 coding conservation; other authors argue at least 7,000 are translated and that counts vary about 30-fold.
-
How much is a buried salt bridge worth?Buried salt bridges are reported as destabilising by a continuum calculation, stabilising by over 4 kcal/mol by experiment, and almost silent for global stability in RNase H*. big-open open-question not yet written not yet written needs-research not yet written partly written
-
How much of a chromatin assay peak set is artefact?CUT&Tag recovers about half of ENCODE ChIP-seq histone peaks, a preprint reports signal unique to CUT&Tag at promoters, and ChIP-seq itself has known artefacts, so no assay is a clean ground truth. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. In K562, CUT&Tag recovered an average 54% of ENCODE H3K27ac and H3K27me3 peaks. A preprint reports 18% of K562 H3K27me3 CUT&Tag peaks have no ChIP overlap. ChIP has its own artefacts, so neither is truth.
-
How much of an imaging-based spatial molecule table is assigned to the wrong cell?Rates of unassigned transcripts and foreign-cell signal exist from the authors of segmentation tools, but no one overall rate of misassigned molecules was found. big-open open-question not yet written not yet written needs-research not yet written partly written
-
How much of curated pathway content do the genome-scale interaction maps recover, and the reverse?How much of curated pathway content do the genome-scale interaction maps recover, and the reverse? big-open open-question not yet written not yet written not yet written not yet written partly written
-
How much of observed alternative splicing makes a protein that does something?How much of observed alternative splicing makes a protein that does something? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. 92 to 94% of human genes are reported alternatively spliced, an extrapolation from short reads. At protein level, one analysis of eight proteomics sets found evidence of any splice event for 246 genes, and 99.63% of detected peptides map to one isoform per gene; deeper proteomics found junction peptides for 34.3% of 13,450 alternative events.
-
How much of the disagreement between pathway databases is definitional rather than curation error?Pathway databases agree on few reactions, and part of the gap comes from where each draws the pathway's boundary; no study separated the two. big-open open-question not yet written not yet written needs-research not yet written partly written
-
How much of the human mRNA-protein gap is measurement rather than biology?Noise-corrected figures exist for yeast and across human tissues, but none for one human cell type with matched RNA and protein replicates. big-open open-question not yet written not yet written not yet written not yet written partly written
-
How much organ-level information does an H&E slide lose?No opened source measures what a single H&E section loses of the organ it came from. amit open-question not yet written not yet written not yet written not yet written partly written
-
How much vitamin C is enough? big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- What intake prevents scurvy, what intake saturates plasma and cells, and what intake is optimal for wound healing and other functions, with the primary studies for each? Published recommendations differ between countries.
-
How noisy is a reaction yield between repeats, and does adding failed reactions help prediction models?No numeric yield noise between repeats was found, and whether failed reactions help yield models is not settled. big-open open-question not yet written not yet written needs-research not yet written partly written
What is measured about yield and conditions?. Yield prediction had mean absolute error 15.6% against 16.3% for guessing the mean, and over half of 15.4 million filtered Reaxys reactions had no yield recorded.
-
How often is context recorded on curated and text-mined edges?The only measured rates found are about one third for SIGNOR and about 63% for text-mined HIPPIE; none exists for INDRA, Biolink or BioGRID. amit open-question not yet written not yet written needs-research not yet written partly written
-
How were the conserved subunits of the 1958 DNA experiment shown to be single polynucleotide chains? big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- The 1958 paper said it had not been shown that the subunits are single chains. Which later experiments settled this, and how? The research chain did not open them.
-
If CAR T cells are the patient own cells, why are there antibodies to them, and why do those antibodies not wipe them out?FDA reviewers reported antibodies against the mouse-derived CAR binder in most tisagenlecleucel patients before infusion, with no apparent effect on expansion or response; no opened source explains why. big-open open-question not yet written not yet written needs-research not yet written partly written
The puzzle. The cells are the patient own, yet antibodies exist against the receptor, and they did not stop expansion or response in the FDA review. Candidate reasons are untested here.
-
In a crowded cell, which partner wins when several proteins compete for one site?In a crowded cell, which partner wins when several proteins compete for one site? amit open-question not yet written not yet written not yet written not yet written partly written
-
In structural bioinformatics, what controlled vocabularies are missing?After a proper search, a database or flat vocabulary exists for most structural concepts and one early ontology covers conformation; pocket, interface as typed entities and an MD metadata standard remain empty. amit open-question not yet written not yet written needs-research not yet written partly written
His question. "In structural bioinformatics, what ontologies are missing, and where is there room to innovate in terms of general ontologies in biology for multiomics?"
-
Is orthology or evolutionary information an edge type anywhere, and does it help prediction?Hetionet carries one evolutionary edge type, rate covariation, but no orthology edge was found in any graph read and none has been tested. big-open open-question not yet written not yet written not yet written not yet written partly written
-
Is the irreversibly silent state of an engineered gene a molecular lock, and how long do silenced states last?A fitted model describes an irreversibly silent state; no study identifies the molecular lock, and the longest measured durations are partial and in mice. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. The best evidence for a point of no return is a fitted three-state model. The longest engineered silencing reported is nearly a year in mice at about a halving. No molecular lock has been identified.
-
Is there a metadata standard for molecular dynamics simulations?Partly: the community has published a call for a standard and a minimum-metadata list, and two databases have their own field lists, but no agreed standard was found on the pages read. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Is there a public, variant-level, expert-curated set of loss-of-function, gain-of-function and dominant-negative labels?No variant-level expert mechanism set was found in the public material opened. big-open open-question not yet written not yet written not yet written not yet written partly written
What is known and what is not?. Mechanism labels exist per gene and disease. A variant-level expert set was not found in the sources opened.
-
The list of open problems in biology he mentions big-open open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- He recalls a recent blog post from Edison Scientific listing open problems in biology "like the Millennium Problems". Find it, read it, and write one record per problem.
-
Two published rates for antibodies against the Kymriah CAR, years apartSources from different years report different rates of antibodies against the mouse-derived part of the tisagenlecleucel CAR; no record says which is right. big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- What rate does each source report, measured how, in which patients, and at what time after infusion?
- Is the difference a real change (assay, population, follow-up) or two ways of counting the same thing?
- Does either rate predict loss of response? Open the primary immunogenicity data, not a review.
-
Was any process-engineering or food-company reaction dataset ever used for a biological model, or was none released?No public process-scale or food-company reaction dataset was found used for a biological model; whether none exists or none was released is unknown. big-open open-question not yet written not yet written needs-research not yet written partly written
What do the public food and process sources hold?. USDA FoodData Central holds composition and manufacturer-supplied nutrient data, with no reaction or kinetic fields on its landing page. The one well-studied shared reaction family, Maillard chemistry, reached biology as ideas and assays.
-
What a selectivity panel cannot see To write it: How much of the proteome does a standard panel cover, and what is systematically outside it (non-kinase targets, states, complexes)? not yet written
To write this, these must be answered:
- How much of the proteome does a standard panel cover, and what is systematically outside it (non-kinase targets, states, complexes)?
- Does residence time at an off-target change the picture a single-concentration panel gives? Two gears flagged by report 4's concept report (free-energy perturbation windows; panel versus kinetic residence-time selectivity) were never run under its gear cap, so this exists only at concept depth.
-
What are the median half-lives of human mRNA and protein, from one cell type by one method?What are the median half-lives of human mRNA and protein, from one cell type by one method? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. Human mRNA half-life: mean 6.9 h in HeLa (an abundant-transcript subset); earlier mammalian medians 40 min to about 9 h; protein: about 10 to over 1,000 h in non-dividing primary cells. No single median was confirmed.
-
What counts as proof that a drug will work? amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- What equations and evidence are accepted as proof at each stage? Can each delivery question be shown in isolation, or only end to end in an animal?
-
What does a term or evidence code connote to a curator beyond its definition?This atlas found curator guidance on how reliable each GO evidence code is, but no study framed as connotation. big-open open-question not yet written not yet written needs-research not yet written partly written
-
What does an edge type connote to a practitioner beyond its definition?Authors' caveats show that edge types like binds or associated_with are read as stronger than they are defined; what a compound or structure node connotes is unanswered. big-open open-question not yet written not yet written not yet written not yet written partly written
-
What drives offset pi stacking: quadrupole electrostatics or dispersion against repulsion?The textbook Hunter-Sanders explanation is disputed; a 2020 critique and a 2025 defence use the same kinds of calculation on isolated dimers. big-open open-question not yet written not yet written needs-research not yet written partly written
-
What fraction of a graph benchmark score survives a degree-only baseline and a by-disease split?Nobody outside the builders has reported how much of a PrimeKG or ogbl-biokg link-prediction score is left after a degree-only baseline and a split by disease. big-open open-question not yet written not yet written not yet written not yet written partly written
-
What fraction of clinical failures could animal models have predicted?No opened source ties clinical failures back to the animal model that preceded them. big-open open-question not yet written not yet written not yet written not yet written partly written
-
What fraction of genes and proteins is lost or split when mapping between Ensembl, RefSeq, HGNC and UniProt?No measured loss rate between the main gene and protein identifier systems was found in any source opened for report 37. big-open open-question not yet written not yet written needs-research not yet written partly written
-
What happens to a fully uncapped nascent transcript in a human cell?What happens to a fully uncapped nascent transcript in a human cell? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. Capping happens while Pol II is paused: the most proximal paused RNAs are largely uncapped and distal ones capped, as a model; the review calls cause and effect 'still unclear'. Capping 'does not always proceed to completion', and the enzyme DXO degrades capped but unmethylated pre-mRNA in mammals. The fate of fully uncapped nascent RNA has no primary citation in the sources opened.
-
What happens to a stored claim in an agent system when a later experiment contradicts it?What happens to a stored claim in an agent system when a later experiment contradicts it? big-open open-question not yet written not yet written not yet written not yet written partly written
What is known. Kosmos updates its world model with summaries of task outputs. The co-scientist appends Meta-review text to later prompts. The AI Scientist-v2 adds debugging children to failed nodes. Kitano's 2021 paper asked for belief revision.
-
What is the real error rate of cross-references and SSSOM mappings?Error rates are measured for samples of mappings but not for the OBO cross-reference network as a whole. big-open open-question not yet written not yet written needs-research not yet written partly written
-
What is the registration error, in micrometres, when spatial modalities are aligned?The only measured micrometre error found is 4.5 micrometres for mass-spectrometry imaging; none was found for Xenium-to-Visium or H&E-to-Xenium. big-open open-question not yet written not yet written needs-research not yet written partly written
-
What is the splicing error rate per intron in a living human cell?What is the splicing error rate per intron in a living human cell? big-open open-question not yet written not yet written needs-research not yet written partly written
What is observed. The best figure is indirect: about 0.7% of reads at junctions with conserved sites go to unconserved sites (75 lymphoblastoid lines, short reads), roughly 2% of transcripts of a median four-exon gene. Novel junctions are 0.32 to 1.08% of junction reads in GTEx v8. Both count only RNAs that survived decay.
-
What should exist in biology and has never been seen? amit expected-unobserved not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- Are there things we would expect to exist that do not, or that exist and should not? For example, proteins that fuse RNAs from two different parts of the genome.
-
What was the last universal common ancestor? big-open open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- What is known, what is not, and what would settle it?
-
What would a CRO or cloud lab charge for protein-interaction assays?What would a CRO or cloud lab charge for protein-interaction assays? amit open-question not yet written not yet written not yet written not yet written partly written
-
When and why did the Navy and merchant marine change from lemon juice to lime juice? big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- Find the primary Admiralty and merchant-shipping records for the change from lemon to lime juice, with dates and reasons, and the evidence on how much weaker lime juice was as an antiscorbutic. The research chain found only secondary accounts with differing dates.
-
Where does a negative result live in an agent system's memory?Where does a negative result live in an agent system's memory? big-open open-question not yet written not yet written not yet written not yet written partly written
What is known. The co-scientist authors say negative results and paywalled prior art are missing from what it reads. In the AI Scientist-v2 a failed experiment is a tree node that is not the best. Robin and Kosmos report what worked.
-
Where is money moving among sequencing, proteomics, organoids, chips and animals?No primary source gave a funding series by modality; only cost per approval and isolated figures. amit open-question not yet written not yet written not yet written not yet written partly written
-
Which chromatin change explains a GWAS signal?eQTLs colocalise with a minority of GWAS signals, chromatin-accessibility QTLs add some loci mostly without a transcript link, and the numbers use different denominators. big-open open-question not yet written not yet written not yet written not yet written partly written
The short version. Reported shares of GWAS signals with a known mechanism range from about 8% to a little over half, depending on assay and denominator. The numbers cannot be put on one scale.
-
Who conceived the density-label experiment on DNA replication, and what did each author contribute? big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- Read the participants' own accounts and the historians (Holmes's histories were not opened) to establish who proposed labelling DNA with a heavy isotope and banding it in a cesium chloride gradient, when, and what Meselson, Stahl and Vinograd each contributed.
-
Who validates language-model-proposed ontology terms when less confident raters accept fluent definitions?Less confident raters scored AI-written definitions as good as human ones, which leaves open who checks them. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Why did revusiran, an siRNA against transthyretin, have more on-treatment deaths than placebo?ENDEAVOUR was stopped for a mortality imbalance, and the authors state that no clear causative mechanism was identified. big-open open-question not yet written not yet written needs-research not yet written partly written
The question in brief. Two drugs against the same protein had opposite outcomes, and the cause of the harm is unknown.
-
Why did tominersen make patients worse while lowering mutant huntingtin?In GENERATION HD1 the antisense drug lowered mutant huntingtin in the spinal fluid, yet the every-2-months arm did worse than placebo and dosing was halted; the cause is not established. big-open open-question not yet written not yet written needs-research not yet written partly written
The lesson in brief. Lowering the target protein is a measurement that the drug works, not that the patient benefits. See drugs/tominersen for the sourced numbers.
-
Why do biomedical ontologies end, and who studies it?Retirements are recorded but their causes mostly are not, and no systematic study was found. big-open open-question not yet written not yet written needs-research not yet written partly written
-
Why do most people infected with Helicobacter pylori never develop an ulcer? big-open open-question not yet written not yet written needs-research not yet written not yet written
To write this, these must be answered:
- What distinguishes the infected people who develop a duodenal or gastric ulcer from those who do not (bacterial strain, host genetics, acid output, smoking, drugs), with the primary studies? The 1984 survey already found the organism in many patients without ulcer.
-
Why do orphan proteins have no known partners or function? amit open-question not yet written not yet written not yet written not yet written not yet written
To write this, these must be answered:
- What is hard about the dark proteome and dark genome, and what has been tried?