Codon Digest: Discovering Antibiotics with Deep Learning
Plus: A protein that edits cell membranes.
Codon Digest is my weekly roundup of research, news, and industry highlights about engineered biology. Please send me your feedback.
Note: My work at MIT and Asimov has picked up significantly (in exciting ways that I’ll write about soon!), so this Digest will be published more irregularly. Thanks for reading.
This week: A way to measure a transgene’s expression in the brain using ultrasound, a DNA sequencing method that uses 1000x less reagents, and base editors get even smaller.
Let’s dive in.
A note from our sponsor:
Our friends at Fifty Years are the best at helping world-class scientists become world-class entrepreneurs. They believe a key to solving the world's biggest problems is helping more scientists translate their work via startups. If you're thinking of starting a startup, reach out to them at codon@fiftyyears.com. Fifty Years is also proud to sponsor Codon.
🔥 Amazing Things
…that happened this week, in 5 minutes or less.
1/ Deep Learning-Guided Antibiotic Discovery
A new antibiotic kills Acinetobacter baumannii, which is one of three superbugs identified by the W.H.O. as a “critical threat” to humanity. It’s called abaucin.
How? Nearly 8,000 molecules were screened in the lab to find those that inhibited the growth of A. baumannii, a microbe that is notorious for its high resistance to off-the-shelf antibiotics. The screened molecules had known chemical structures, and nearly 500 of them were active against the bacteria. These data were then used to train a message-passing neural network to “predict” which other chemical structures might also inhibit growth of A. baumannii.
The neural network was unleashed on a database of 6,680 compounds with unknown activities and produced a “shortlist” of possible antibiotics in 1.5 hours. Out of 240 molecules on the shortlist that were tested in the lab, 9 inhibited A. baumannii by at least 80%. Abaucin, the winning molecule, was also validated in mice and human samples. It’s highly specific against A. baumannii and doesn’t kill much else.
This new molecule — unlike traditional antibiotics — targets the lipoprotein trafficking system. It interferes with a protein complex, called LolE, that shuttles lipoproteins from the inner to the outer membrane. A single mutation at amino acid position 362 in this protein, though, confers resistance to abaucin.
So What? This paper is amazing, but not unprecedented. Other studies have similarly applied machine learning to antibiotic discovery, including a paper last year that identified antimicrobial peptides in the human gut using deep learning. But this paper seems far more general, and I like that they delved so deep into the mechanism of how the chemical works. I’m assuming it’s now possible to screen lots of molecules against nearly any microbe, and run this neural network to find other narrow-spectrum candidates with high activity. That seems profoundly useful.
Read more in Nature Chemical Biology.
2/ Shrinking Base Editors
IscB is a weird gene-editing enzyme that has low activity in human cells. An engineered version of this protein can convert DNA bases with efficiencies up to 92%. It is, in other words, a new type of miniature base editor.
How? The researchers first compared the editing efficiency of different versions of IscB when coupled with 'ωRNA,' which guides the enzyme to the right spot on the DNA. A particular variant, named IscB*-ωRNA*, had the highest editing efficiency across multiple different sites in the genome.
The protein was fused to an exonuclease, an enzyme that cuts DNA strands. The resulting fusion protein was tested at 23 sites across five genes, and had editing efficiencies up to 22 times higher than wild-type IscB, reaching levels of 61%.
Some versions of IscB can also be converted into base editors. By mutating just three amino acids in the wildtype protein (D61, E193 and H247), the authors made a nickase (it ‘nicks’ DNA, but does not cut it). Then, they generated two types of base editors: One that converts A to G, and another that converts C to T, with editing efficiencies up to 63% and 66%, respectively.
So what? This paper is a big deal! The wild-type IscB protein only has 500 amino acids. Base editors made from Cas9 are about twice as big. This means that IscB base editors might be easier to package inside of AAVs, which can only store 4,700 bases of DNA for gene therapies.
Read more in Nature Methods.
3/ Cell Membrane Editor
A cell is a lipid shell with lots of DNA, proteins, and RNA molecules inside. If I sit down, close my eyes, and think about a cell, I suppose it’d be obvious to think that there must be proteins that ‘tweak’ the shell, convert one lipid into another, make new lipids, and so on. But never, in my three years of writing this newsletter, do I recall hearing about someone who engineered a cell membrane-editing protein to make new types of lipids.
For a new study, researchers engineered an enzyme called phospolipase D (which breaks phosphatidylcholine into a lipid and choline molecule) to be 100-fold more active than its wildtype version, and also to make lipids with designer ‘heads’.
How? Researchers introduced between 5 and 11 mutations in the PLD enzyme to make a series of superPLD variants. Each variant was screened using an in vitro assay.
The resulting superPLDs can catalyze two different reactions: Hydrolysis and transphosphatidylation (exchange the head groups of phospholipids), which means it’s now possible to make a whole slew of natural and unnatural phospholipids with designer chemical groups.
The superPLD enzymes became “fast” thanks to a single structural tweak: A histidine at position 440 shifted into a new position, made a larger catalytic pocket, and could thus more easily take in phospholipid substrates.
So what? This paper is just cool. The engineered enzymes can edit a cell’s membrane to change its stiffness, or to turn living cells into “designer phospholipid” factories. It opens up a whole new class of biological engineering. Cell engineers are often so focused on DNA, RNA and proteins, while neglecting anything related to lipids. Hopefully this paper changes that.
Read more in Nature Chemistry.
🧪 From the Lab
Other wet-lab papers worth checking out. (* = Recommended)
*Sequencing by avidity enables high accuracy with low reagent consumption. Nature Biotechnology. Read
A DNA sequencing method that saves money by reducing some required reagents by 1,000x. It’s also super accurate.
*Light-driven biosynthesis of volatile, unstable and photosensitive chemicals from CO2. Nature Synthesis. Read
Some bacteria can convert CO2 to other molecules, but they’re difficult to engineer. This paper presents a modular approach to convert CO2 into olefins and other molecules by assigning different tasks to distinct microbes.
Biosynthesis of cannabigerol and cannabigerolic acid, the gateways to further cannabinoid production. Synthetic Biology. Read
This is fun (but not unprecedented). By taking enzymes from Cannabis sativa and repurposing them, it was possible to synthesize cannabigerols in E. coli cells.
Development of a chemogenetic approach to manipulate intracellular pH. JACS. Read
A new method to control pH inside a living cell using engineered enzymes and pulses of light.
*Spontaneously established syntrophic yeast communities improve bioproduction. Nature Chemical Biology. Read
*Acoustically-targeted measurement of transgene expression in the brain. bioRxiv. Read
*An engineered influenza virus to deliver antigens for lung cancer vaccination. Nature Biotechnology. Read
*Stoichiometric expression of messenger polycistrons by eukaryotic ribosomes (SEMPER) for compact, ratio-tunable multi-gene expression from single mRNAs. bioRxiv. Read
*Generation of a mutator parasite to drive resistome discovery in Plasmodium falciparum. Nature Communications. Read
Strand-selective base editing of human mitochondrial DNA using mitoBEs. Nature Biotechnology. Read
Meningococcal ACWYX conjugate vaccine in 2-to-29-year-olds in Mali and Gambia. NEJM. Read
A fully integrated wearable ultrasound system to monitor deep tissues in moving subjects. Nature Biotechnology. Read
Walking naturally after spinal cord injury using a brain–spine interface. Nature. Read | News
Improving the soluble expression of difficult-to-express proteins in prokaryotic expression system via protein engineering and synthetic biology strategies. Metabolic Engineering. Read
A non-transmissible live attenuated SARS-CoV-2 vaccine. Molecular Therapy. Read
Synthetic virology approaches to improve the safety and efficacy of oncolytic virus therapies. Nature Communications. Read
Programming inactive RNA-binding small molecules into bioactive degraders. Nature. Read
Synthetic biology pathway to nucleoside triphosphates for expanded genetic alphabets. ACS Synthetic Biology. Read
Engineered bacteria to accelerate wound healing: an adaptive, randomised, double-blind, placebo-controlled, first-in-human phase 1 trial. eClinicalMedicine. Read
Gene expression dynamics in input-responsive engineered living materials programmed for bioproduction. Materials Today Bio. Read
Living microecological hydrogels for wound healing. Science Advances. Read
Maximizing protein production by keeping cells at optimal secretory stress levels using real-time control approaches. Nature Communications. Read
Modular metabolic engineering and synthetic coculture strategies for the production of aromatic compounds in yeast. ACS Synthetic Biology. Read
Spatial imaging of glycoRNA in single cells with ARPLA. Nature Biotechnology. Read
Genetically encoded photocatalytic protein labeling enables spatially-resolved profiling of intracellular proteome. Nature Communications. Read
An RNA origami robot that traps and releases a fluorescent aptamer. bioRxiv. Read
Cultured meat platform developed through the structuring of edible microcarrier-derived microtissues with oleogel-based fat substitute. Nature Communications. Read
A pilot study of closed-loop neuromodulation for treatment-resistant post-traumatic stress disorder. Nature Communications. Read
Intracellular RNA and DNA tracking by uridine-rich internal loop tagging with fluorogenic bPNA. Nature Communications. Read
A noncommutative combinatorial protein logic circuit controls cell orientation in nanoenvironments. Science Advances. Read
💾 Computers x Bio
Papers from the worlds of AI & programming. (* = Recommended)
*An all-atom protein generative model. bioRxiv. Read
*Global analysis of the yeast knockout phenome. Science Advances. Read
*MISATO - Machine learning dataset of protein-ligand complexes for structure-based drug discovery. bioRxiv. Read
Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nature Biotechnology. Read
Inference of cell type-specific gene regulatory networks on cell lineages from single cell omic datasets. Nature Communications. Read
FLUXestimator: a webserver for predicting metabolic flux and variations using transcriptomics data. Nucleic Acids Research. Read
ExpressAnalyst: A unified platform for RNA-sequencing analysis in non-model species. Nature Communications. Read
📈 Data Brief
One chart about biology and our world.
Malaria deaths plummeted between 2010 and 2015, thanks mostly to the billions of dollars in philanthropic funding that were used to purchase bed nets and medicines. Most deaths are in children under 5 years, and the first vaccine to get endorsed by the WHO for “broad use” in this age group was Mosquirix…in 2021.
Only about 1.5 million children have received this vaccine so far, and the rollout is far from simple. Infants need at least three doses before age 2, and the vaccine only reduces severe cases by about 30%. The good news, though, is that many more vaccines are on the way. R21, developed at the University of Oxford, was up to 80% effective in a small clinical trial of 450 children in Burkina Faso.
Mosquitoes kill more humans, by far, than any other animal. Combatting malaria, dengue, and other diseases will be the great medical achievement of our generation. We are living through history.
📤 In the News
Rapid-fire news items you might have missed.
Why is it so difficult to make gene-edited rice? Well, it doesn’t help that a single experiment can take 3+ months! My essay for Works in Progress magazine. Read
Asimov, a company that builds tools to program living cells, launched an engineering blog. I’m writing some of them and I hope you’ll subscribe. Read
Heartcore Capital released their AI & Productivity Report, which touches on how AI is shaping software, biotech, and some other fields. I wrote a blurb for it. Read
People with brain implants, who later have them removed, feel worse off. As Neuralink, Paradromics and other brain:machine interface companies raise money and put implants into people, we’ll hear much more about this issue. MIT Technology Review. Read | Read
A tablet version of semaglutide, the weight loss drug, helped people shed 15.1% of their body weight (compared to 2.4% on placebo) over 17 months. No more injections. STAT. Read
We’re super close to making eggs or sperm by reprogramming other cells in the body. It’d be a big deal for IVF. NPR. Read
A curated list of AI resources from a16z. Recommend. Read
The F.D.A. needs more time to decide whether or not to approve Sarepta’s gene therapy for Duchenne muscular dystrophy. (An advisory panel recommended approval in a contentious 8-6 vote.) Fierce Biotech. Read
U.S. lab in Kansas will work with deadly animal pathogens. Science. Read
Neuralink says it has F.D.A. ‘OK’ to start clinical trials. Ars Technica. Read
Opentrons is releasing a new line of AI-compatible pipetting robots. Read
GenScript says they can synthesize more challenging DNA sequences between 200 and 1,800 base pairs in length. Read
Synthace now integrates ChatGPT with their lab automation platforms. Scientists can “use a natural language interface to describe their intent, use the AI to design complex experiments, and then run those experiments on lab equipment.” Read | Video
HuidaGene Therapeutics, a Chinese gene-editing company, says they have developed the first DNA base editor that can convert guanine to cytosine or thymine. Labiotech. Read
Novartis bought a blood stem cell gene therapy program for $87.5 million in cash. Labiotech. Read
VarmX raised €30M to run a clinical trial for blood clotting. They make a drug, called VMX-C001, that is a recombinant form of blood clotting factor X protein. Labiotech. Read
ReNAgade Therapeutics raised a $300M Series A. The RNA therapeutics space is flourishing in a time of otherwise tight finances. STAT. Read
The next day, ElevateBio raised $401M. They make gene editors. Fierce Biotech. Read
🧠 Musings
Fun stuff that has little to do with biotech.
ChainForge is a visual environment for prompt engineering. Recommend. Read
An excellent essay about Thomas Edison, and why he was brilliant despite the criticisms. “He took tech that barely worked and made it useful. He took products to market at prices people were willing to pay. He made things that could be manufactured at scale. He hit deadlines.” By Eric Gilliam for Works in Progress. Read
“Graduate students report anxiety and depression at rates six times that of the general population,” according to data from 2,300 respondents in 25 countries. The numbers are worse for women. Nature. Read
Remember that famous picture of Ben Franklin holding a kite during a thunderstorm? Yeah, it’s all made up. It shows Franklin’s son as a young boy, but he would have been 21 at the time. And Franklin did the experiment while standing under shelter. Ars Technica. Read
Nanoparticles are now the world’s lightest paint; no pigments needed. Ars Technica. Read
Compostable plastic cups are often made from polylactic acid, or PLA. They’re touted as being ‘green,’ but don’t break down in seawater even after 14 months. New Scientist. Read
A device turns breast milk into freeze-dried powder. Tech Crunch. Read
⚒️ In the Ether
Education, resources, and events.
The BioxML hackathon is happening this week. The demos are this Sunday, June 4th! View more
Fifty 50 is a free program for PhDs and postdocs who want to start a company. (And I’m pretty sure you can nominate yourself.) Apply
A new game studio is making “Minecraft for synthetic biology.” They’re looking for developers. Read
👀 Meme of the Week
“I booked the FACS from 1pm” @OdedRechavi
Until next time,
— Niko
Disclosure: The views expressed in this blog are entirely my own and do not represent the views of any company or university with which I am affiliated.
Really cool about using deep learning to discover a new antibiotic! I wonder when/if it'll be possible to simulate human clinical trials at scale, too, and then take the most promising results to real-human trials. You'd have to simulate entire human bodies and all the possible interactions of molecules and effects, but someday....