Codon: Notes on Progress #1


AI breaks from proteins. Neurotech ignition. Bio+tech papers from the last month.

“Biology is the only engineering domain where we can physically measure thousands, millions, and sometimes billions of design[s] simultaneously in real experiments.”
Gleb Kuznetsov

Last month, I killed off the weekly Index. Its remains now lie buried in email inboxes, cursed to an eternity of nothingness. But as I put down the shovel — what’s that? — faint whispers reach my ears.

The weekly research posts were useful! You should bring them back!

At first, I try to tune out the voices. Must resist urge to spend my weekends skimming through research papers! But the voices grow louder. And now, after weeks spent indecisively tottering back and forth between ‘agree’ and ‘deny,’ I’ve made a compromise: A monthly Codon Index. This is part one. It’s quite long and dense. If you have ideas for how I can spruce it up and make it ✨more fun✨ to read, please let me know!

AI for Proteins Everything Else

Some scientific breakthroughs take the world by storm. Others never leave the ethereal chambers of a college campus. We’re living through a peculiar era, though, in which more science is translating into the public domain than ever before. In the context of biology, most people know that AI tools can design never-before-seen proteins, while other models have basically flown under the radar. (I know this because my relatives, most of whom are attorneys, jet mechanics, or philosophers, have asked me about protein design.) It’s great to hype up David Baker and OpenAI every other week, but I don’t really think they need the attention. Deep learning models are useful for more than just amino acids.

For one recent study, a high-throughput experiment was first used to measure the gene editing outcomes — both intended and unintended — for more than 92,000 guide RNAs. The authors then used these data to train a neural network that could predict editing efficiencies for new guide RNAs. The model’s accuracy was far higher than prediction tools, from just two years ago, that were designed to do the same thing!

Other studies, in recent weeks, have also used machine learning to discover new antimicrobial peptides, design biological circuits with desired properties, or predict how metabolisms evolve over millions of years. For that last paper, a machine learning model, trained on just 3,000 bacterial genome sequences, predicted which genes are gained and lost from an organism at various branches in their evolutionary tree. The results suggest “that evolutionary pressures and constraints on metabolic systems are universally shared.” And the tool, called Evodictor, may help scientists forecast “the future evolution of genomes.”

Elsewhere, scientists are merging machine learning and automation to design and execute experiments in much the same vein that I discussed in my January essay with Alexander Titus, called “AI for Science: Dreams of Progress.” Amyris, a synthetic biology company in Emeryville, California, developed Lila, an ‘Automated Scientist,’ in early January. Lila designs metabolic pathways that, when inserted into a living cell, can produce a specific chemical. Lila automatically selects which genes to insert into the cells, and then tests everything on a robot. Out of 454 target molecules, Lila successfully overproduced 242. Not bad.

Neurotech’s Ignition

A brain-computer interface, developed by Stanford University scientists, enabled a patient with ALS to speak a record 62 words per minute, which is 3.4 times faster than the prior record. The BCI measures neural patterns in the cortex, using microelectrodes, and has an error rate of 9.1 percent for each word. The co-senior author, Krishna Shenoy, has published more than 150 papers over the last thirty years. A giant in neuroscience and neural recording technologies, he died shortly after this study was published.

In other BCI-related news, safety data was reported for fourteen adults who had been implanted with a microelectrode array over a 17-year period, stretching from 2004 to 2021. During that time, there were just 6 serious adverse events, none of which were serious enough to warrant device removal. The fully implanted devices lasted for 872 days, on average, which is quite impressive.

And, finally, Max Hodak left Neuralink, the company he co-founded with Elon Musk, in May 2021 to start his own neural prosthesis venture, called Science.XYZ. His team has already released a preprint describing their first product: A thin LED implant to stimulate vision.

Rabbits were first injected (in the eyeball) with adeno-associated viruses, or AAVs, that carry genes encoding a light-sensitive opsin protein. These viruses are designed to target specifically retinal ganglion cells, which connect the retina to the brain’s visual cortex. Animals are then implanted with a 30 micron-thick LED device (2.5-times thinner than a sheet of paper) that flashes lights directly onto the retina, thus activating the retinal ganglion cells with high resolution. The team successfully triggered “activity in the contralateral visual cortex. This approach suggests a path to an implanted optogenetic therapy that operates at cellular resolution.”

Photos of the device. Knudsen et al. in bioRxiv.

The List 🔻

AI + Software

†An automated scientist to design and optimize microbial strains for the industrial production of small molecules. Singh et al. bioRxiv. Link

†Predicting prime editing efficiency and product purity by deep learning. Mathis et al. Nature Biotechnology. Link

Prime editing is a CRISPR-based gene editing tool that can add small pieces of DNA to the genome, delete DNA, or swap individual bases. This study analyzed the editing outcomes of 92,423 pegRNAs targeted against 13,349 human mutations and then built a neural network to reliably predict prime editing outcomes. This is a useful dataset and tool to improve gene-editing.

†Identification of potent antimicrobial peptides via a machine-learning pipeline that mines the entire space of peptide sequences. Huang et al. Nature Biomedical Engineering. Link

Genome sequence databases are swelling in size. Trillions of bases, and billions of DNA sequences, have been catalogued for more than half-a-million species. Useful antimicrobials and other medicines are hiding within those data. A new machine-learning pipeline discovered 14 new antimicrobial peptides after screening through 512 billion sequences in just 27 days.

Meta learning improves robustness and performance in machine learning-guided protein engineering. Minot & Reddy. bioRxiv. Link

Machine learning enables prediction of metabolic system evolution in bacteria. Konno & Iwasaki. Science Advances. Link

Machine learning for optimization of multiscale biological circuits. Merzbacher et al. bioRxiv. Link

Signal peptide efficiency: From high-throughput data to prediction and explanation. Grasso et al. ACS Synthetic Biology. Link

The Nucleotide Transformer: Building and evaluating robust foundation models for human genomics. Dalla-Torre et al. bioRxiv. Link

Generative power of a protein language model trained on multiple sequence alignments. Sgarbossa et al. eLife. Link

Undersampling and the inference of coevolution in proteins. Kleeorin et al. Cell Systems. Link

Structure-based prediction of T cell receptor:peptide-MHC interactions. eLife. Link

Basic Research

†Histone-organized chromatin in bacteria. Hocher et al. bioRxiv. Link

Histones are big proteins that tightly coil DNA. They help all six feet of the human genome pack inside each and every cell. It was thought, for decades, that only eukaryotes have histones. But no; some bacteria also have histones, and they use them to compact DNA in a slightly different way (they form a shell around the genetic material, rather than coil it up).

Metabolic regulation of species-specific developmental rates. Diaz-Cuadros et al. Nature. Link

Visualization of translation and protein biogenesis at the ER membrane. Gemmer et al. Nature. Link

Minimal synthetic enhancers reveal control of the probability of transcriptional engagement and its timing by a morphogen gradient. Alamos S. et al. Cell Systems. Link

A DNA methylation atlas of normal human cell types. Loyfer et al. Nature. Link

Promoter sequence and architecture determine expression variability and confer robustness to genetic variants. Einarsson et al. eLife. Link


†A high-performance speech neuroprosthesis. Willett et al. bioRxiv. Link

Brain-computer interfaces have roughly doubled in capabilities over the last three years. This study shows a “speech-to-text” device that records neural data from the cortex and enables people with ALS to speak more than 60 words per minute, with a 9.1% error rate at each word.

†Acoustically targeted noninvasive gene therapy in large brain regions. Nouraein et al. bioRxiv. Link

Pulses of focused ultrasound can briefly open up holes in the blood-brain barrier, thus enabling medicines to be delivered into the central nervous system. This study used sound to open 105 barrier sites in mice, and then delivered AAVs for gene therapies with very high transduction efficiencies. One step closer to delivering any medicine into the brain; immune system be damned.

†Assessment of safety of a fully implanted endovascular Brain-Computer Interface for severe paralysis in 4 patients. Mitchell et al. JAMA Neurology. Link

Thomas Oxley, an Australian neurologist, first began developing the Stentrode, a small electrode array that sits in a blood vessel (rather than near neurons), back in 2010. The FDA granted Synchron, Oxley’s company, breakthrough status in August 2020. Stentrodes are implanted via stents, rather than open-skull surgery, and record neural data much like a conventional microelectrode array. Early safety data, on a small number of patients, is now in: No serious adverse events, and the devices last 12+ months. Synchron beat Neuralink to human trials, but they also had a hefty head start.

Droplet-based transcriptome profiling of individual synapses. Niu et al. Nature Biotechnology. Link

Near-infrared fluorescence lifetime imaging of amyloid-β aggregates and tau fibrils through the intact skull of mice. Hou et al. Nature Biomedical Engineering. Link

A hardware system for real time decoding of in vivo calcium imaging data. Chen et al. eLife. Link

An optimized bioluminescent substrate for non-invasive imaging in the brain. Su et al. Nature Chemical Biology. Link

Robust and adjustable dynamic scattering compensation for high-precision deep tissue optogenetics. Li et al. Communications Biology. Link

Acoustic trapping and navigation of microrobots in the mouse brain vasculature. Fonseca et al. bioRxiv. Link

An updated suite of viral vectors for in vivo calcium imaging using intracerebral and retro-orbital injections in male mice. Grødem et al. Nature Communications. Link

Interim safety profile from the feasibility study of the BrainGate neural interface system. Rubin et al. Neurology. Link

Engineered Cells

In vivo expression vector derived from anhydrobiotic tardigrade genome enables live imaging in Eutardigrada. Tanaka et al. PNAS. Link

A genetic toolkit modifies the DNA of living tardigrades, which live for a staggering 30 years without food or water. The toolkit includes a plasmid and some genes that are known to work in water bears.

Engineering a scalable and orthogonal platform for synthetic communication in mammalian cells. Pistikou et al. bioRxiv. Link

CRISPR-associated type V proteins as a tool for controlling mRNA stability in S. cerevisiae synthetic gene circuits. Yu & Marchisio. Nucleic Acids Research. Link

Short tRNA anticodon stem and mutant eRF1 allow stop codon reassignment. Kachale et al. Nature. Link


†Engineered live bacteria suppress Pseudomonas aeruginosa infection in mouse lung and dissolve endotracheal-tube biofilms. Mazzoloni et al. Nature Biotechnology. Link

Mycoplasma pneumoniae bacteria, which normally cause minor lung infections, were engineered to express genes that help kill pathogens and break down biofilms. These engineered microbes were able to “dissolve biofilms formed in endotracheal tubes of patients with ventilator-associated pneumonia.” Turning nature against itself!

†Shark nanobodies with potent SARS-CoV-2 neutralizing activity and broad sarbecovirus reactivity. Chen et al. Nature Communications. Link

Nurse sharks that have been immunized with the SARS-CoV-2 receptor binding domain produce antibodies that have a “high affinity to all…viral variants of concern, including the divergent Omicron strains.” If you’re wondering how sharks actually get immunized, then you’ve come to the right place — I had the same thought. Six sharks “were held in a continuously-recirculating 12,000 L seawater tank maintained at 28 °C” in Baltimore, Maryland. They were injected through their lateral fins, presumably by scuba divers? Blood was collected two weeks after immunization.

Transcriptional reprogramming restores UBE3A brain-wide and rescues behavioral phenotypes in an Angelman syndrome mouse model. O’Geen et al. Molecular Therapy. Link

Large-scale differentiation of iPSC-derived motor neurons from ALS and control subjects. Workman et al. Neuron. Link

Multiplex epigenome editing of MECP2 to rescue Rett syndrome neurons. Qian et al. Science Translational Medicine. Link

Assessment of systemic AAV-microdystrophin gene therapy in the GRMD model of Duchenne muscular dystrophy. Birch et al. Science Translational Medicine. Link

Synapse-tuned CARs enhance immune cell anti-tumor activity. Chockley et al. Nature Biotechnology. Link

Therapeutic adenine base editing of human hematopoietic stem cells. Liao et al. Nature Communications. Link

Bacterial expression of a designed single-chain IL-10 prevents severe lung inflammation. Montero-Blay et al. Molecular Systems Biology. Link

Displaying and delivering viral membrane antigens via WW domain–activated extracellular vesicles. Choi et al. Science Advances. Link

Design of a pan-betacoronavirus vaccine candidate through a phylogenetically informed approach. Lewitus et al. Science Advances. Link

Peptide-guided lipid nanoparticles deliver mRNA to the neural retina of rodents and nonhuman primates. Herrera-Barrera et al. Science Advances. Link

Protein & Molecular Engineering

†Large language models generate functional protein sequences across diverse families. Madani et al. Nature Biotechnology. Link

ProGen is a deep learning tool that generates protein sequences with a desired function. It was trained on 280 million protein sequences. When it was used to make new-to-Earth antibacterial proteins, many effective sequences were generated that had low homologies to each other (usually less than 20%), which is pretty amazing. It seems that ProGen can design proteins that have the same functions but look completely different from one another.

†A universal deep-learning model for zinc finger design enables transcription factor reprogramming. Ichikawa et al. Nature Biotechnology. Link

ZFDesign is another deep-learning model, trained on 49 billion protein-DNA interactions, that can be used to make new transcription factors with desired target sequences. Again, this is amazing, and will vastly expand the genetic circuits that can be built in living cells. The authors made new-to-Earth repressors and activators, with tunable activities, to control genes.

Combinatorial assembly and design of enzymes. Lipsh-Sokolik et al. Science. Link

Unlocking de novo antibody design with generative artificial intelligence. Shanehsazzadeh et al. bioRxiv. Link

Fast, accurate ranking of engineered proteins by receptor binding propensity using structural modeling. Ding et al. bioRxiv. Link

ProT-VAE: Protein transformer variational autoencoder for functional protein design. Sevgen et al. bioRxiv. Link

Tension-tuned receptors for synthetic mechanotransduction and intercellular force detection. Sloas et al. Nature Biotechnology. Link

Reading and Writing DNA

†Simultaneous sequencing of genetic and epigenetic bases in DNA. Füllgrabe et al. Nature Biotechnology. Link

A useful technique sequences DNA and tracks specific DNA methylation markers, which are typically used to measure the biological age of an organism, in a single experiment.

†A highly specific CRISPR-Cas12j nuclease enables allele-specific genome editing. Wang et al. Science Advances. Link

A ‘new’ types of CRISPR-Cas protein, called Cas12j, can edit one allele in the genome while leaving the other untouched. This study screened six different Cas12j proteins and showed that one variant, Cas12j-8, successfully disrupted four individual target loci without causing off-target edits at other alleles.

Recovering false negatives in CRISPR fitness screens with JLOE. Dede & Hart. Nucleic Acids Research. Link

If one uses CRISPR gene-editing to knock out hundreds of genes across a genome to find, say, cancer-associated risk factors, about 20 percent of all hits will be ‘false negatives.’ This method recovers those false hits. Nucleic Acids Research

Modeling CRISPR-Cas13d on-target and off-target effects using machine learning approaches. Cheng et al. Nature Communications. Link

Improved cytosine base editors generated from TadA variants. Lam et al. Nature Biotechnology. Link

Genome editing with natural and engineered CjCas9 orthologs. Gao et al. Molecular Therapy. Link

Programmable RNA detection with CRISPR-Cas12a. Rananaware et al. bioRxiv. Link

Metabolically-targeted dCas9 expression in bacteria. Pellegrino et al. Nucleic Acids Research. Link

TadA orthologs enable both cytosine and adenine editing of base editors. Zhang et al. Nature Communications. Link

Digital data storage on DNA tape using CRISPR base editors. Sadremomtaz et al. bioRxiv. Link

TadA reprogramming to generate potent miniature base editors with high precision. Zhang et al. Nature Communications. Link

Selection of extended CRISPR RNAs with enhanced targeting and specificity. Herring-Nicholas et al. bioRxiv. Link

RNA recording in single bacterial cells using reprogrammed tracrRNAs. Jiao et al. Nature Biotechnology. Link

Enhancement of a prime editing system via optimal recruitment of the pioneer transcription factor P65. Chen et al. Nature Communications. Link

The impact of nucleosome structure on CRISPR/Cas9 fidelity. Handelmann et al. Nucleic Acids Research. Link

Massively parallel knock-in engineering of human T cells. Dai X. et al. Nature Biotechnology. Link

Precise transcript targeting by CRISPR-Csm complexes. Colognoi et al. Nature Biotechnology. Link

Passer, a highly active transposon from a fish genome, as a potential new robust genetic manipulation tool. Wang et al. Nucleic Acids Research. Link

Evolution of CRISPR-associated endonucleases as inferred from resurrected proteins. Alonso-Lerma et al. Nature Microbiology. Link

Genome mining for unknown–unknown natural products. Yee et al. Nature Chemical Biology. Link


Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Waziry et al. Nature Aging. Link

A large trial in 220 patients shows that caloric restriction does “not lead to significant changes in biological age estimates” as measured by DNA methylation patterns across the genome.

Gene therapy mediated partial reprogramming extends lifespan and reverses age-related changes in aged mice. Macip et al. bioRxiv. Link

This paper got a ton of attention because it showed that body-wide expression of three genes — OCT4, SOX2 and KLF4 — extends the remaining lifespan of 124-week-old mice by 109 percent. Whether this will ever translate to humans is uncertain, to say the least.

The antitumour effects of caloric restriction are mediated by the gut microbiome. Mao et al. Nature Metabolism. Link

Photoactivatable senolysis with single-cell resolution delays aging. Shi et al. Nature Aging. Link

Optogenetic rejuvenation of mitochondrial membrane potential extends C. elegans lifespan. Berry et al. Nature Aging. Link

Making a Cell

Robust and tunable performance of a cell-free biosensor encapsulated in lipid vesicles. Boyd M.A. et al. Science Advances. Link

Light-activated assembly of connexon nanopores in synthetic cells. Sihorwala et al. JACS. Link

Science & Society

†National Institutes of Health research project grant inflation 1998 to 2021. Lauer et al. eLife. Link

Nearly half of all NIH grant money now goes to solicited research projects (a.k.a. there is less money to do truly ‘blue skies’ or ‘exploratory’ research.) The NIH is also funding more clinical trials and more projects over $5M than ever before. And, of course, the discrepancy between the top 1% of projects and "the rest" of grantees has grown over time. Still, I’m really glad that the NIH chose to release these data, which are a valuable starting point for policy reforms and debates.

Analysis of science journalism reveals gender and regional disparities in coverage. Davidson & Greene. eLife. Link

The Wild & Plants

†Heritable transgene-free genome editing in plants by grafting of wild-type shoots to transgenic donor rootstocks. Yang et al. Nature Biotechnology. Link

The tools available to engineer plants are really bad. This study presents a new way to engineer plants, using CRISPR, while removing the transgenes in a single generation. It could possibly speed up plant breeding in certain crops by several years.

†Genome editing in plants using the compact editor CasΦ. Li et al. PNAS. Link

Two small Cas9 proteins show “much higher editing efficiency” in plants, relative to wild-type proteins.

Engineering a symbiont as a biosensor for the honey bee gut environment. Chhun et al. bioRxiv. Link

A natural biological adhesive from snail mucus for wound repair. Deng et al. Nature Communications. Link

Closing the gap to effective gene drive in Aedes aegypti by exploiting germline regulatory elements. Anderson et al. Nature Communications. Link

Haploid male fertility is restored by parallel spindle genes in Arabidopsis thaliana. Aboobucker et al. Nature Plants. Link

A female in vivo haploid-induction system via mutagenesis of egg cell-specific peptidases. Zhang et al. Molecular Plant. Link

Tools & Technology

†Massively parallel protein-protein interaction measurement by sequencing (MP3-seq) enables rapid screening of protein heterodimers. Baryshev et al. bioRxiv. Link

This study is my favorite in this entire list. A new (quite complicated) method can quantitatively measure the strength of protein-protein interactions, over several orders of magnitude, using DNA sequencing. The authors show that the method “can scale to measure over 100,000 interactions at once.”

†Unidirectional single-file transport of full-length proteins through a nanopore. Yu et al. Nature Biotechnology. Link

A nanopore protein was used to sequence full-length proteins with roughly 90 percent accuracy at each amino acid position. The method works by sensing slight fluctuations in an electrical current. Proteins are unfolded using a denaturing chemical, called guanidinium chloride, and the tool reads about 100 amino acids per second.

A pseudovirus system enables deep mutational scanning of the full SARS-CoV-2 spike. Dadonaite et al. Cell. Link

Video-based pooled screening yields improved far-red genetically encoded voltage indicators. Tian et al. Nature Methods. Link

Spatial transcriptomics for profiling the tropism of viral vectors in tissues. Jang et al. Nature Biotechnology. Link

Drone-assisted collection of environmental DNA from tree branches for biodiversity monitoring. Aucone et al. Science Robotics. Link

Remote control of muscle-driven miniature robots with battery-free wireless optoelectronics. Kim et al. Science Robotics. Link

Other Interesting Things

Severe multi-year drought coincident with Hittite collapse around 1198–1196 BC. Manning et al. Nature. Link

The Hittite empire, under Muršili II, straddled the land between the Black and Mediterranean seas. It stretched from modern-day Istanbul, in the west, to Syria in the east. This powerful nation collapsed around 1200 B.C., a time that oddly coincides with “an unusually severe continuous dry period” that lasted three years and was unraveled, oddly, by studying tree rings in the region’s juniper trees. An amazing paper.

Gender inequality and self-publication are common among academic editors. Liu et al. Nature Human Behavior. Link

An intriguing study on self-publications amongst journal editors. A full 6 percent of editors, out of 20,000 analyzed, “published at least one-third” of their papers in their own journal.

Mixed yeast communities contribute to regionally distinct wine attributes. Hawkins et al. FEMS Yeast Research. Link

Why does a Pinot Noir from Napa Valley taste different from another in Oregon? Sure, the climates and soils are different (and play a role), but hyper-regional yeast communities probably explain about ten percent of variations.

Early human impact on lake cyanobacteria revealed by a Holocene record of sedimentary ancient DNA. Nwosu et al. Communications Biology. Link

At-home, cell-free synthetic biology education modules for transcriptional regulation and environmental water quality monitoring. Jung et al. bioRxiv. Link

Musings on the Future

Tweets, essays & articles that made me think.
  • Colossal, the de-extinction company that says they’ll bring back a woolly mammoth (or, erm, an “Arctic Elephant”), raised $150M in a Series B. I’ll have more to say about this very soon.
  • LanzaTech, a company that uses engineered microbes to convert factory emissions into ethanol and other chemicals, is now trading on the NASDAQ. This seems like a genuinely big deal for synthetic biology, and the company is saying that they’re the first “carbon capture and transformation” company to go public in the U.S.
  • Sam Rodriques, a professor at the Crick Institute in London, shares his visions for biology’s future. The two that resonated with me: “Can we create a full molecular model of an individual human cell?” and “Can we build an AI scientist that can do experiments and make inferences about the world as fast or faster than we can?” I’ve partly covered both of these topics, here and here.
  • Answer open-ended research questions (kinda like Elicit), but with *multiple* citations in one text response. Looks like another useful, AI-powered tool for research. And the code was released on GitHub!
  • A new Army Center for Synthetic Biology is hiring a program manager.
  • An article in Nature claims that disruptive science has dropped, precipitously, since 1945 — a full 90 percent, to be exact. But George Church and Juergen Eckhardt had some issues with the methods used to make that claim, and I tend to agree with them. Biological engineering has been soaring in the last two decades, even if we haven’t **yet** achieved synthetic biology’s grand vision of truly “programmable cells.”
  • Ashton Trotman-Grant, the former Bio Lead at San Francisco-based VC, Fifty Years, has been writing an excellent newsletter about the intersection of biotechnology and video games.
  • On January 13, I tweeted about that famous DNA polymerase video (y’know, the one shown in high school classes) and the original animator replied! Drew Berry made the video in 2003. Biology needs more animators and artists, like David Goodsell and Janet Iwasa.
  • Microsoft released BioGPT, a generative model that “trained on large-scale biomedical literature” and can help scientists quickly scan through published experiments and data. The paper was actually released in September, but I missed it.
  • I spent an entire week talking to as many DAOs as I possibly could, including LabDAO and ValleyDAO. But I still feel, most of the time, like I have no clue what they’re talking about. Molecule DAO’s biotech arm released a helpful, free book, called the “BioDAO Bible,” which has resolved some of my confusion.
  • I’m really tired of all the ChatGPT tweets, but the model is surprisingly good at explaining biology through metaphors or analogy. This is the first time I’ve considered using GPT to write very small sections of my articles.
  • For the last three months, I’ve been researching bioengineering curriculums as part of my work at MIT. A shocking discovery: Many engineers are required to take statistics courses, but those courses overwhelmingly use clinical trials or epidemiology studies as their case studies. It is rare for bioscientists to take a statistics course that emphasizes bioscience experiments and examples. That’s why I was so happy to see Wolfgang Huber and Susan Holmes drop a **free* textbook online, “Modern Statistics for Modern Biology,” with examples in R.
  • When companies spin out from academic labs in the U.K., their host universities take more than 3x the equity of U.S. universities. That’s embarrassing, really.
  • is a free website, developed by Dwarkesh Patel and Ryan McWhorter, to search through thousands of books. Ask a question, set a time range, and get results. When I asked, “How did the Roman, Seneca, die?” the tool’s first result was Tacitus, with the correct answer. “…he went into a pool of hot water, spattering the slaves closest to him…he was then taken into the bath, where he suffocated in the steam. He was cremated with no funeral ceremony.”
  • A graduate student rubbed spicy sauce on obese mice for 8 weeks.

You’re a legend. Thanks for reading.
— Niko McCarty
(Email | Twitter)