July 2021 SubCriticalAppraisal Newsletter
With links on polygenic embryo screening, mirror DNA polymerase, AlphaFold2, Temple of Artemis, and more.
This is the inaugural July 2021 issue of the SubCriticalAppraisal Newsletter (archives).
The purpose of this newsletter is to be a more meaningful and summarised version of the RSS feed, complete with all my writings, site updates, interesting links, and reviews each month.
Feel free to email me with any and all suggestions or feedback. Without further ado:
The first Alzheimer's drug in 18 year – aducanumab (Aduhelm™), a monthly (for 18 months) IV beta-amyloid IgG monoclonal antibody, but it doesn't actually work – it didn't show reduction in clinical decline, only reduction in beta-amyloid levels. Aduhelm was approved on biomarker endpoints, not primary efficacy endpoints; it's like targeting bilirubin to treat hepatitis.
Their two stage 3 trials were halted early because of futility. They then did some post hoc analysis and found the first stage 3 trial's interim data on the high dose regimen showed a statistically significant difference of -22% in clinical decline derived from a difference of 0.39 on an 18-point score of cognitive and functional skills. Other cognitive test endpoints showed no difference e.g. MMSE difference was 0.6 points (not even the minimal clinically significant difference of 1). No statistically significant reduction in clinical decline in lower dose in that trial or any dose in the second stage 3 trial.
10 out of 11 panellists on the FDA advisory panel ultimately voted that the presented data could not be considered as evidence of aducanumab’s effectiveness; the remaining panellist was uncertain. Then the FDA approved it anyway, buying into the amyloid hypothesis although every single attempt to translate that into a beneficial clinical effect has failed – this meta-analysis combined results from 14 randomized controlled trials and found that reduction in amyloid levels alone is unlikely to substantially slow cognitive decline within the follow-up period of most typical trials. This shows regulatory inconsistency, and misdirects future pharma R&D towards a likely dead end. In reversal, the FDA now suggests that only those with mild memory or thinking problems should receive it.
The Institute for Clinical and Economic Review estimates that a cost-effective price is $2,500–8,300 per year, and the cost per QALY threshold by NICE for England and Wales is between £20,000 and £30,000. Aduhelm costs $56,000/year not counting the cost of routine PET scans to monitor the common (41% in treatment vs 10% in placebo!) side effects of cerebral edema/ sulcal effusions/ microhaemorrhages/ superficial siderosis, so who’s paying? The vast majority of US patients who take it will have Medicare coverage of some kind, so the cost will be largely borne by taxpayers, so the libertarian argument doesn’t apply because patients wouldn’t be taking Aduhelm on their own nickel.
First polygenically-screened baby born, see also Embryo Screening for Polygenic Disease Risk: Recent Advances and Ethical Considerations, Tellier at al 2021: New overview of state-of-the-art polygenic risk scores, precision genotyping of embryos, and genomic indices; and Preimplantation Genetic Testing for Polygenic Disease Relative Risk Reduction: Evaluation of Genomic Index Performance in 11,883 Adult Sibling Pairs, Treff et al. 2020:
Embryo Selection for Intelligence: Gwern estimates that with 10 embryos (a near-best-case scenario of what you're likely to get from egg extraction) and modern (as of 2016/2017) polygenic scoring technology, you could get on average +3 IQ points by implanting the smartest; if polygenic scoring technology reached the limits of its potential (might happen within a decade or two) you could get +9 IQ points
Related: The Algernon Argument – any simple major enhancement to human intelligence is a net evolutionary disadvantage.
Intelligence to us looks like an unalloyed good but just like everything in biology it's a trade off; there ain't no freed lunch and intelligence is subject to diminishing returns like every other thing; in effect all animals are under stringent selection pressure to be as stupid as they can getting away with.
Exhibit A: Ashkenazi jew theory of intelligence – the Ashkenazi were forced into occupations demanding intelligence, and micro-selected for high intelligence; the trade-off tho is that too much of the high IQ genes result in horrible diseases like Tay-Sachs.
Exhibit B: a Scottish family had a unique mutation which was found to increase verbal IQ in afflicted family members vs non by something like 25 points, except that mutation starts causing blindness in their 20s.
CRISPR-Cas9 In Vivo Gene Editing for Transthyretin Amyloidosis, Gillmore et al. 2021: We really are finally on the cusp of a brave new CRISPR world of medical therapeutics (limited to a few rare germline genetic diseases will be amenable to in vivo gene-editing therapy via IV infusion like this one).
Efficacy of Wolbachia-Infected Mosquito Deployments for the Control of Dengue, Utarini et al. 2021: Remember all the doomsaying about releasing engineered mosquitoes into the wild? Well the authors just released Aedes aegypti mosquitoes infected with the wMel strain of Wolbachia pipientis (less susceptible than wild-type to dengue virus infection) in 24 geographic clusters, and found protective efficacies (n=8144) of 77.1% for symptomatic dengue and 86.2% for hospitalisation.
“What’s with the Hungarian supergenius scientists?” Drug discovery edition: Tamas Bartfai co-created the first acellular pertussis (whooping cough) vaccine; co-created the “coolmouse” thus disproving the dogma that all mammals have a 36.7 °C core body temperature; co-discovered the muscarinic acetylcholine receptor in the brain which was thought to be only present in the periphery; and co-discovered the coexistence of classical transmitters and neuropeptides; directed the development of the first SSRI (zimelidine) for depression, the first proton pump inhibitor (omeprazol) for acid reflux, the COMT inhibitor (Tasmar-tolcapone) for Parkinson’s disease, the first benzodiazepine antagonist (flunitrazepam) for benzodiazepine overdose, and the first oral multiple sclerosis drug (sphingosine 1-phosphate).
A guide to cancer immunotherapy: from T cell basic science to clinical practice, Waldman et al. 2020
Rapid Sequencing-Based Diagnosis of Thiamine Metabolism Dysfunction Syndrome, Owen et al. 2021: Brave new world of rapid sequencing-based diagnoses: the genome of an infant with encephalopathy was sequenced in just over 11 hours, which led to a clinical diagnosis of thiamine metabolism dysfunction syndrome 2 (THMD2), informing the treatment of the child, without which would lead to infant death.
Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans, Roberts et al. 2021: all 400+ chest X-ray COVID detection ML algorithms published since COVID are found to be fatally flawed and thus have zero clinical use. Problems include using a dataset consisting of paediatric patients below the age of 5 as the control, using duplicate datasets, implicit bias in the COVID positive x-ray scans, and the many blatant methodological errors include failures to reveal key details about the training and experimental data sets, not performing robustness or sensitivity analyses of their models, not performing any external validation, not showing any confidence intervals around the final results etc.
Accuracy of Practitioner Estimates of Probability of Diagnosis Before and After Testing, Morgan et al. 2021: survey of 553 GP staff found widespread overestimates (by 2-10 times) of the probability of diseases; adjustment of estimates in face of test results ranged from accurate to yet more overestimation. Med schools need to teach pretest probability estimates! Teach the predictive values of patient history, examination, and investigations! Teach Bayesian clinical reasoning, and provide readily accessible references for clinical sensitivity and specificity, make pretest probability calculators for diseases other than cardiac ischemia!
Richard Smith, former editor of the BMJ: Time to assume that health research is fraudulent until proven otherwise? It’s worse than you thought.
Likelihood of Null Effects of Large NHLBI Clinical Trials Has Increased over Time, Kaplan et al. 2015:
Sarah Constantin’s Green Flags for Biomedical Results
In a mad world, all blogging are psychiatry blogging:
Mapping neural circuit biotypes to symptoms and behavioral dimensions of depression and anxiety, Goldstein-Piekarski et al. 2021
Freddie deBoer’s tragic account of his suffering in the US mental healthcare system.
A 16-year-old boy fell into a 3-metre-deep whole wheat grain wagon, and his lungs were so filled with grain that they could not collapse in spite of oxygen mask ventilation. He could only be intubated after enough grain was removed from his mouth and pharynx using suction and forceps. It took the thoracic surgeons 4 hours to remove all visible grains by clapping on his chest and suction via bronchoscope. He had a PaCO2 of 501 mmHg, the highest ever reported (normally 38-42 mmHg). The boy made a full recovery and was discharged 10 days after admission.
Human muscle stem cells are refractory to aging, Novak et al. 2021
Carbohydrates, insulin, and obesity, Speakman et al. 2021
COVID: How did we do? How can we know? Our regulatory state is still failing us, moronically and tragically. Related: Broad cross-national public support for accelerated COVID-19 vaccine trial designs, Broockman 2021: Survey (n=5920) across 8 countries found broad majorities prefer for scientists to conduct challenge trials (75%) and integrated trials (63%) over standard trials.
Lessons From The Crisis: The transatlantic institutional anti-mask campaign, summarised, Things can get really bad, really fast, way faster than regular people will notice, Was Ebola a near-miss? The experts can stay wrong longer than you can stay alive. Substack by the same person behind the harrowing Covid One Year Ago (@YearCovid)
Mass mask-wearing notably reduces COVID-19 transmission, Leech et al. 2021: (n=20M) an entire population wearing masks in public leads to a median reduction in R of 25.8%, though it has nothing to do with mandating mask-wearing. See also: Normalizing Community Mask-Wearing: A Cluster Randomized Trial in Bangladesh, Abaluck et al. 2021
Lockdown Effectiveness: Much More Than You Wanted To Know and Things I Learned Writing The Lockdown Post: Is it even possible to know? The biggest factors in evaluating policy cost-effectiveness are often not even mentioned in the Discourse e.g. “emotional damage” and “Long COVID” (what even is it?)
The Origins of SARS-CoV-2: A Critical Review, Holmes et al. 2021; The proximal origin of SARS-CoV-2, Andersen et al. 2020; Early appearance of two distinct genomic lineages of SARS-CoV-2 in different Wuhan wildlife markets suggests SARS-CoV-2 has a natural origin, Garry et al. 2021; Spike protein sequences of Cambodian, Thai and Japanese bat sarbecoviruses provide insights into the natural evolution of the Receptor Binding Domain and S1/S2 cleavage site, Holmes et al. 2021
CoVariants: Graphs show, for each country, the proportion of total number of sequences (not cases), over time, that fall into defined variant groups.
Cov2Tree: Interactive SARS-CoV-2 phylogeny.
CORONA GAME: Play as the Czech Republic in a simulation of how COVID-19 spreads, game ends when either enough citizens become immune or the public gets so unhappy that your government falls.
Bioorthogonal information storage in l-DNA with a high-fidelity mirror-image Pfu DNA polymerase, Fan et al. 2021: straight out of a Arthur C. Clark short story – mirror-image DNA polymerase has been made, capable of stable data storage and gene assembly, along with L-chiral DNA-dependent RNA polymerase. Next step is to make a ribosome but the authors need to make a L-DNase otherwise nothing can get rid of the L-DNA once it’s made. New x-risk just dropped: mirror life version of cyanobacteria that only needs achiral nutrients and light for photosynthesis could take over Earth's ecosystem due to lack of natural enemies, destroying the bottom of the food chain by producing mirror versions of the required sugars (with the upside of solving global warming).
The AlphaFold2 papers are here: methods Highly accurate protein structure prediction with AlphaFold, Jumper et al. 2021; application of predictions Highly accurate protein structure prediction for the human proteome, Tunyasuvunakool et al. 2021; Open source code for AlphaFold; AlphaFold Protein Structure Database; Explainer of one of AF2’s secret sauces: equivariant structure prediction; Ongoing Open Source effort to reproduce AF2 in PyTorch; from Yours Truly: Did DeepMind Solve The Protein Folding Problem? And Solution To Levinthal's Paradox
The complete sequence of a human genome, Nurk et al. 2021; and A complete reference genome improves analysis of human genetic variation, Aganezov et al. 2021; see also Why it took 20 years to 'finish' the human genome — and why there's still more to do
Infinite re-reading of single proteins at single-amino-acid resolution using nanopore sequencing, Brinkerhoff et al. 2021
Inferring a Continuous Distribution of Atom Coordinates from Cryo-EM Images using VAEs, Rosenbaum et al. 2021
RNA demethylation increases the yield and biomass of rice and potato plants in field trials, Yu et al. 2021: transgenic expression of the human RNA demethylase FTO in rice causes more than threefold(!) increase in grain yield under greenhouse conditions, ~50% increases in rice and potato yield and biomass in field trials.
Biologically indeterminate yet ordered promiscuous gene expression in single medullary thymic epithelial cells, Dhalla et al. 2019: Thymic epithelial cells express almost the entire protein coding genome! And maintain epithelial cell identity!
A connectomic study of a petascale fragment of human cerebral cortex, Shapson-Coe et al. 2021: The 1.4 petabyte image is viewable online.
Retinotopic organization of visual cortex in human infants, Ellis et al. 2021
Untangling the cortico-thalamo-cortical loop: cellular pieces of a knotty circuit puzzle, Shepherd et al. 2021
Life, uh, finds a way:
Biology is more theoretical than physics, Gunawardena 2013: the enzyme-substrate complex, the receptor, the ion channel, the gene, and the tumour suppressor were all hypothesised to exists, and mathematical reasoning used to show their explanatory power for experimental findings, for decades before they were discovered to actually exist.
'Infotaxis' as a strategy for searching without gradients, Vergassola et al. 2007
Detecting Causality in Complex Ecosystems, Sugihara et al. 2012
2.4 Physics/Computer Science
Elastic ice microfibers, Xu et al. 2021
Rechargeable self-assembled droplet microswimmers driven by surface phase transitions, Cholakova et al. 2021 (video)
Deep learning with coherent nanophotonic circuits, Shen et al. 2017
Quantum phases of matter on a 256-atom programmable quantum simulator, Ebadi et al. 2021
The Modern Mathematics of Deep Learning, Berner et al. 2021
The Bayesian Learning Rule, Khan et al. 2021
FNet: Mixing Tokens with Fourier Transforms, Lee-Thorp et al. 2021
Lossy Compression for Lossless Prediction, Dubois et al. 2021
A visual introduction to Gaussian Belief Propagation, Ortiz et al. 2021
Scale invariant robot behavior with fractals, Kriegman et al. 2021
Andrej Karpathy’s Recipe for Training Neural Networks
New overview of the replication crisis in Psychology: Replicability, Robustness, and Reproducibility in Psychological Science, Nosek et al. 2021
Creative Destruction: The Structural Consequences of Scientific Curation, McMahan et al. 2021
Stuart Ritchie on How the Lancet lost our trust
Andrej Karpathy’s Survival Guide to a PhD
Beginning in the late sixties, Helmut Kentler, one of the most influential sexologists in Germany, placed neglected boys in foster homes run by pedophiles. The "Kentler experiment" was authorized and financially supported by the Berlin Senate. In a report submitted to the Senate, in 1988, Kentler had described it as a “complete success.”
The Cape Town–Port Elizabeth Railway service officially employed Jack, a chacma baboon, as a signalman, and paid it twenty cents a day, and half a bottle of beer each week. It is widely reported that in his nine years of employment with the railway company, Jack never made a single mistake.
People systematically overlook subtractive changes, Adams et al. 2021
The wisdom of the inner crowd in three large natural experiments, Dolder et al. 2017
Bullshit Ability as an Honest Signal of Intelligence, Turpin et al. 2021
Kids these days: Why the youth of today seem lacking, Protzko et al. 2019
Retraction Note: Complex societies precede moralizing gods throughout world history, Whitehouse et al.: props to the authors for retracting their 2019 paper that made a big splash.
Montreal’s night of terror on 7 October 1969 “cured Steven Pinker of his adolescent anarchism”.
Facebook Victory Against the FTC Is an Antitrust Lesson In Monopoly Power: “Hipster antitrust” – even if Facebook cannot raise prices for consumers i.e. it doesn’t meet the minimum legal definition of monopoly, FB is evil so the antitrust laws are wrong.
GitHub Copilot: suggests whole lines or entire functions in your editor, powered by OpenAI’s Codex, a GPT language model fine-tuned on publicly available code from GitHub, more at Research recitation - GitHub Copilot, and Evaluating Large Language Models Trained on Code, Chen et al. 2021
Andrej Karpathy’s from-scratch tour of Bitcoin in Python
Can Individual Investors Beat the Market? Coval et al. 2021
What's your (sur)name? Intergenerational mobility over six centuries, Barone et al.. 2016
Pseudoerasmus on Why the Service Sector was Historically the Key to Development
Byrne Hobart on The Economics of Pseudonymity
Existential risk from AI and orthogonality: Can we have it both ways? Müller et al. 2021:
Premise 1: Superintelligent (general) AI is a realistic prospect and would be out of human control (Singularity claim)
Premise 2: Any level of intelligence (instrumental) can go with any goals (Orthogonality thesis)
Conclusion: Superintelligent (general and instrumental) AI poses an existential risk for humanity
80000 Hours condensed their decade of research and advice into a free 8-week course, then compressed that into Planning a high-impact career: a summary of everything you need to know in 7 points
Holden Karnofsky: All Possible Views About Humanity's Future Are Wild
One of the Seven Wonders of the World is canonically the best. Antipater of Sidon, who wrote the list used today, said when he laid eyes on the Temple of Artemis at Ephesus, “Those other marvels lost their brilliancy, and I said, ‘Lo, apart from Olympus, the Sun never looked on aught so grand.’” Artemis is the goddess of childbirth, and her temple was burned down the day Alexander the Great was born – legend has it that Artemis was too preoccupied with Alexander’s delivery to save her burning temple. Alexander must have felt guilty because he offered to pay for the temple's rebuilding when he conquered Ephesus. The Ephesians tactfully refused, saying "it would be improper for one god to build a temple to another" and eventually rebuilt it after Alexander’s death, at their own expense.
In 1942, a British armed merchant ship was torpedoed by a German U-boat off the coast of Brazil. As the ship was sinking, Chinese second stewart Poon Lim took a life jacket and jumped overboard before the boilers exploded. When a large storm hit which spoiled his fish and fouled his water, Lim, barely alive, caught a bird and drank its blood to survive. When there was no rain, he caught a shark, clubbed it to death using water jugs, cut it open, and sucked the blood from its liver. He survived on his life raft for 133 days, a record still unbroken today.
Kids are people too: constant lying, casual dishonesty, blatant ageism, candy-coated contempt, patronising baby talk, and content censorship; echoes my own experience as a summer camp counselor.
Paul Graham on How to Work Hard
Robin Hanson on Experts Versus Elites
Sarah Constantin on How To Be An Educated Layman
Tyler Cowen on How do you ask good questions?
Byrne Hobart on Writing: Good Career Move, Terrible Career; How Much Money Do Authors Actually Earn?; On Being a Loser, or Should you Start a Blog?; and Tyler Cowen’s Advice for possible and wanna-be book writers
The narcissistic fall of France: Could’ve only been written by Michel Houellebecq.
Pocket Guide to China: A 64 page primer for American soldiers stationed in China during World War II, a charming time capsule of old-school liberalism, propaganda, and clunky-yet-earnest cultural tolerance.
Wei Dai’s tale from Communist China
The New York Times: Fields of Watermelons Found On Mars, Police Say