Tag Archives: science

AI art is fun

I don’t have much to say on the AI art public debate yet, other than to say it feels overly vitriolic. But I can definitely say that AI art has been fun to make. I’d love to have the time to put some of these together into a 1-off RPG campaign, but I don’t know when I’d have the time. I’d also like to use some time to learn how I can use an AI art program on my own computer and train it against specific images I want, rather than relying on the web-browser based programs that are trained on a whole host of unnecessary data. 

But that’s for another time. For now, AI art is fun to toy around with, and dream of what could be very soon. 

Not feeling good, but hope to have a post tomorrow

As the title says. Work is stressful when you’re not sure if you’re looking in the right direction. We have a lot of seemingly contradictory data coming in from our experiments, but those contradictions are hopefully pushing us in the direction of new science. Just how the shape of proteins relates to their disease states is still new ground, and I’m proud to be working in it. But it’s a difficult field to get data in.

I sometimes look back and wish I could have worked in an earlier time. You look at say Gregor Mendel’s pea plants and you think that that would have been a scientific endeavor you could do easily. Or discovering new elements in the 19th century when all you needed was an atomic mass and the mass ratios of an oxygen and fluorine salt. I think back and that research seems so much easier, since high school kids replicate some of those experiments today.

But I know it isn’t so simple. It’s easy to replicate those things first because we know what we’re looking for, and second because our technology is so much better. Finding the structure of benzene is a classic case, students learn how to draw benzene very early on in an organic chemistry class, but its structure confounded the best and brightest for decades in the 19th century. They didn’t have the accuracy of scales we do, the easy access to light-based detection methods, or nuclear based detection methods, hell they didn’t even have a theory of protons and neutrons. They knew Carbon and Hydrogen existed, but they weren’t in any way sure how to fit those onto a periodic table yet, and weren’t certain there wasn’t some new element hiding around in Benzene. Putting together an experiment to prove the structure of benzene, using only 19th century knowledge and 19th century technology, is a lot harder than it sounds, and me wishing I could have worked on that discovery instead of my current one is “grass-is-greener-ism.”

Another good one for any discussion: we often laugh at those silly medievals who believed the sun goes around the earth. I mean, even some Greek philosophers proposed it, but alas the medievals were just too closed-minded, right? But actually the geocentric theory did seem to be parsimonious for a good long while. Here’s a fun thought experiment: how would you prove geocentricism using only what you could find in the 10th century? No telescope, no pictures from orbit, just observations of the sky. If you know your astronomy you know there are certain irregularities with the orbits of planets as viewed from earth, and that is a good argument against geocentricism. Yet it was also noted that there is no perception of movement when one is standing on the earth, and that was taken as an argument against heliocentricism. It wasn’t until Galileo’s theory of relative motion that a cogent counter-argument was put in place, and so if you want to prove heliocentricism in the 10th century you’d also have to do the hard work of demonstrating relativity like Galileo did. Copernicus’s model of heliocentricism is often seen as revolutionary, but it still had endless epicycles needed to explain it, more even than geocentricism, making it not that much better than Ptolemy’s geocentricism, so if you want to argue for heliocentricism by attacking epicycles you’d also need to do the hard math that Kepler did in establishing how orbits can be calculated based on ellipses. It really isn’t as easy a problem as it sounds.

So yeah, work is hard but I guess it’s always been hard. We think all the easy discoveries have been made, but those discoveries were made when they were hard to make.

The danger of small patterns

As I’ve probably said before, I work as a researcher. When you’re doing difficult or expensive research, you don’t usually have the time or money to do a whole lot of replications. That goes doubly if you’re working with patients or patient samples. But since science is all about finding patterns, how can you find patterns in a small dataset?

There are statistical tools that can help with this, but even before you get to the hypothesis testing phase, you need to know which direction your hypothesis will go in. For that, we tend to look at the small patterns which aren’t yet statistically significant and try to see what they mean. The danger here is when you don’t get data in a reasonable amount of time, you want to work on your project but you don’t have data to work on. So you go back to whatever you have, the “small patterns” and start extrapolating from there. “If this pattern holds, what could it mean for this disease?”

Then you can start getting attached to a hypothesis that has no data to back it. When you do get data, you may start to interpret it in light of the small pattern you already detected, a pattern which may not even hold. That’s the problem with small patterns, you get to thinking they mean more than they do.

The human brain is a pattern matching machine. Our first calendars came about from noticing that the seasons of a year came in patterns, and that certain stars in the sky could be seen during the hot season while others could be seen during the colder one. But people also thought they detected patterns about how certain things happened on earth when certain stars were seen in the sky. One pattern between stars and the sky held true, there is a correlation between which stars you can see and the season in your local area. But another pattern was false. Yet both patterns were studied and believed for thousands of years.

I hope I don’t get attached to bad patterns for quite so long as that, but it’s hard to avoid. When you’ve got all the time in the world and not enough data, you get attached to these small patterns that you think you detect. And that can hold true even when the pattern is no longer real.

Energy Return on Energy Investment, a very silly concept

Today I’d like to address one concept that I read about in Richard Heinberg’s The End of Growth, Energy Return on Energy Investment or EROEI. The concept is an attempt to quantify the efficiency of a given energy source, and in the hands of Heinberg and other degrowthers it is a way to “prove” that we are running out of usable energy.

EROEI is a simple and intuitive concept, taking the amount of energy produced by a given source and dividing by the amount of energy it costs to set up and use that source. Oil is a prime example. In the beginning of the 20th century oil extract was easy since it just seeped out of the ground in many places. Drilling a small oil well won’t cost you that much, hell you can probably do it with manpower alone. In that case the oil gushing forth will easily give you a good energy return.

In the 21st century however, things have become harder. Oil wells require powerful machines to drill (which costs energy), and the amount and quality of the oi you get out is often lower. Add to that the fact that modern wells require huge amounts of metal and plastics, all of which cost energy to produce and even more energy to transport to their location, then add the energy it took to find the oil wells in the first place using complex geographical surveys and seismographic data, and taken together some people claim that the EROEI for a modern oil well is already less than 1, meaning that more energy is being put in than the energy we get out.

And oil isn’t the only fuel source heading towards and EROEI of less than 1. Modern mining techniques for coal require bigger and bigger machines, natural gas requires more and more expansive facilities, even solar panels require minerals that are more and more difficult to acquire. It seems everything but hydro power and (perhaps) nuclear power are becoming harder and harder to produce, sending energy returns down further and further.

This phenomenon, where the EROEI for our energy sources is less than 1, is supposed to presage an acute energy crisis and the economic cataclysm that degrowth advocates have been warning us about. If we’re getting out less energy than we’re putting in, then we’re really not even gaining, aren’t we? The problem is, I’m struggling to see how EROEI is even a meaningful way to look at this.

First let me note that not all energy is created equal. Energy in certain forms is more usable to us than in others. A hydroelectric dam holds water which (due to its being elevated above its natural resting place) acts as a store of potential energy. The release of that water drives a turbine to produce electricity. But you can’t fly a plane using water power nor keep it plugged in during flight. Jet fuel is another source of potential energy, and it has a number of advantages versus elevated water. Jet fuel is very easy to use and transport, you can fill a tank with it and move it to wherever your plane is, then fill the plane’s tanks from there.

If the only two energy sources in the world were jet fuel and hydroelectric power, we would still find it beneficial to somehow produce jet fuel using hydroelectric power even though that would necessity an EROEI of less than one. Because although this conversion would have less total energy, the energy would be in a more useful form. People would happily extract oil using hydroelectric power, then run refineries using hydroelectric power, because jet fuel has so much utility. This utility means that (supply being equal), jet fuel would command a higher price than hydroelectric power per unit of energy. And so the economic advantages would make the EROEI disadvantages meaningless.

This is the fatal flaw of EROEI in my mind. The fact that some forms of energy are more useful than others means we can’t directly compare energy out and energy in. The energy that is used to run a modern oil well comes to it from the grid, which is usually powered by coal, solar, wind, or nuclear, none of which can be used to fuel a plane. Converting these forms of energy into oil is an economic gain even if it is an energy loss. Furthermore EROEI estimates are generally overly complex and try to account for every joule of energy used in extraction, even when those calculations don’t really make sense. Let me give you an example:

A neolithic farmer has to plow his own fields, sow his own seeds, reap his own corns. Not only that, but the sun’s rays must shine upon his fields enough to let them grow. Billions of kilocalories of energy are hitting his plants every second, and most of then are lost during the plants’ growth process because photosynthesis is actually not all that efficient to begin with. The plant will have used billions of kilocalories of energy, and from them the farmer gets a few thousands of kilocalories of energy. Most of the energy is lost.

This is the kind of counting EROEI tries to do, applied to farming. When you count up every joule of energy that went into the farmer’s food, you find his food will necessarily provide him with an EROEI of less than one thanks to the first law of thermodynamics. But this isn’t a problem because Earth isn’t a closed system, nor are our oil wells. We are blasted by sunlight every minute, our core produces energy from decaying nucleotides, our tides are driven in part by the moon’s gravity, there is so much energy hitting us that we could fuel the entire world for a thousand years and never run out. The problem is that there are some scenarios where that energy isn’t useful. You can’t fly a plane with solar or geothermal or gravitational energy, but you can power an oil well. So we happily use the energies we have lots of (including our use of solar power to grow useful plants and animals!) and use that energy to help us extract the energies with greater utility.

I think EROEI failed from the very beginning for this very reason. It ignores economic realities and the massive amount of energy that surrounds us, and instead argues from the first law of thermodynamics. Yes in any closed system energy eventually runs out, but it isn’t even clear that our universe is a closed system, and the earth definitely is not, so we need to face up to economic reality on this.

10x Genomics: what happened to all the DNA stocks?

2020 was the year of COVID, but it was also the year of DNA. Thousands of DNA companies and researchers got in on the pandemic since there was a sudden surge in demand for COVID testing, contact tracing, and virus studying. Now, COVID is an RNA virus, but RNA and DNA work so similarly that most organizations that do one can do both. Even the Universities got into it, I know the Genomics Research Core Facility at the university I used to work for got money from city, state, and federal governments to process COVID tests, which was way more profitable for them than sequencing my plasmids once a month.

So 10x Genomics was a DNA sequencing company that, like so many others, skyrocketed in valuation during 2020. It hit its absolute peak in early 2021 and then fell precipitously through 2022. That describes a lot of companies but it especially seems to describe DNA companies. Still, I wanted to know if 10x Genomics was a buy, and considering my LinkedIn account got spammed all last year with recruiters and ads for the company, I figured they were at least worth a look. The result? A resounding “eh.”

One of 10x Genomics’ big claims to fame is their ability to perform reads of long segments of DNA as opposed to the shorter segments read by rival Illumina. Sequencing DNA gets less accurate the longer the DNA is, so Illumina and others use a technique of chopping the DNA into pieces, sequencing each piece, and putting them all back together. This usually works fine because there are enough overlapping pieces to make the puzzle fit, if you know read 3 different pieces with letters “AATT”, “TTGG” and “GGCC” then you can hazard a guess that the full sequence reads “AATTGGCC” and that the 3 pieces simply overlapped each other. This doesn’t work with long sequences of a single letter or repeating patters. If you simply have “AAAA” “AAAA” “AAAA” then you actually don’t know how long that string of As is. Those 3 reads could overlap on 2 letter and the result could be “AAAAAAAA” or they could overlap completely and the result is simply “AAAA.” About 8% of the human genome is these sorts of repeating patterns that Illumina and others are ill-equipped to read, which gives 10x an exploitable niche.

Now this is a genuinely interesting piece of equipment and their barcoding of DNA segments to read them in the correct order is a nice bit of chemistry, but was this a company that was ever worth 21 billion dollars? In my opinion *no*. The COVID era was a bubble in a number of ways, the easy-money policies of the post Financial Crisis era led to the super-duper-easy-money policies of the COVID era. Operation Warp Speed and other pandemic-focused funding sources meant that money was flowing into the biotech sector, and with most of the world still coming out of lockdown investors were desperate to park their money in companies that were still able to operate. It seemed like the moment for Biotech had come, and companies like 10x Genomics rode the wave to the very top. But from where I’m standing this was always an obvious bubble, and people were making claims about DNA companies that were woefully unfounded.

I’ve just written a lot about 10x Genomics, but all this could apply to most any DNA/RNA startup around. They all had neat ideas, got woefully overvalued during the pandemic, and have since crashed to earth taking shareholder value along with them. My question today then is why, why does it seem like Wall Street Investors saw something in DNA/RNA companies that I, a biology researcher, never did? Now part of that is that I’m just curmudgeony by nature, but part of that is that I think a lot of investors have this Sci-Fi idea of genetics in their head that isn’t really reflected in the field. I could just point to Cathie Wood and say “she doesn’t know Jack” but I want to dig a little deeper into some of the strange narratives that surround nucleic acids.

First, “DNA as a coding language.” I wonder if computer scientists just latch onto the word “code” or something, because the genetic code is no where near to being something that we can manipulate like computer code, and probably won’t be for decades. We’ve been inserting novel DNA into organisms since the 70s, but recently there’s been a spate of investors and analysts who believe that we’re on the cusp of truly programming cells, being able to manipulate them into doing everything we want as cleanly and as easily as a computer. This would definitely unlock a whole host of industries, it’s also not going to happen for a long while yet. DNA codes for proteins, and proteins are the functional units of most biological processes. Nothing you change in the DNA matters until it shows up in the proteins, and we are far, FAR from being able to understand how to manipulate proteins as easily as we do computers. You cannot simply say “let’s add a gene to make this wheat crop use less water,” you have to find a protein from another plant that causes it to use less water, insert that gene into wheat, do tests to make sure the wheat crop tolerates the new protein, alter the protein to account for unexpected cross reactions, and then finally test your finished wheat product out to make sure it works as designed. Any one of these steps could require an entire company to do, and each step could prove impossible and bankrupt the company working on it. We can’t just make de novo proteins to do our bidding because we don’t even know yet how to predict what a protein will look like when we code for it. Folding at home and other machine learning projects have helped us get partway there, but it will still take many Nobel Prizes before we can make de novo proteins as easily as we make de novo programs. So while there is a genomics revolution going on, it’s still an expensive and time consuming one, it’s not going to solve all our problems in a single go.

Second, “DNA as a storage medium.” I’ve said before that while DNA does store and transmit information, that information cannot be well integrated with our modern technology. The readout of DNA information is in RNA and proteins, while the readout of the circuits in your computer is photons on a screen or electrons in a modem. RNA and proteins do not easily produce or absorb electrons and photons, so having DNA communicate with our current technology is not currently doable. In addition, the time lag between reading DNA and making RNA/proteins is astronomical compared to the speed of information retrieval in semiconductors. I sometimes get an annoyed by the seconds-long delay it takes to load a webpage, but I’d be tearing my hair out if I had to wait on the minutes-long delay for DNA to be transcribed into RNA! At this time I really don’t see any reason to use DNA as a storage medium and I certainly don’t see a path to profit for any company trying to use it as such.

Third, curing genetic diseases. There is definitely a market for curing genetic diseases, let me just say that first, but many of the hyped-up corporate solutions are not feasible and rely more on sci-fi than actual science. I’ve discussed how even though CRISPR can change a cell’s DNA, bringing the CRISPR and cell together is much more challenging. The human body has a lot of defenses to protect itself from exogenous DNA and proteins, and getting around those defenses is a challenge. But in addition I don’t think investors realize that DNA is not the end-all be-all of genetic diseases, and so they tack things on to the Total Addressable Market (TAM) of DNA companies that shouldn’t really be there. Then they get flummoxed when the company has no path to addressing its TAM. Valuing companies based on what they can’t do is a bad investment strategy that I see over and over again with DNA companies. As to genetic disease themselves: there’s a truism in biology that “you are what your proteins are”. Although DNA codes for those proteins, once they’re coded they act all on their own, and some of their actions cannot readily be undone. When a body is growing and developing, its proteins can act up in ways that cause permanent alterations, and after they’ve done so changing the DNA won’t change things back. There are a number of genetic-linked diseases which are not amenable to CRISPR treatment because by the time the disease is discovered the damage has been done and changing the DNA won’t undo the damage.

Finally, “move fast and break things” doesn’t work with DNA the way it does with computers. I’ve worked on both coding projects and wet-lab projects. When something goes wrong in my computer code, fixing it is a long and arduous process but I have tools available that let me know what exactly the code is doing every step of the way. I can step through the code line by line and find out exactly what went wrong, and use that knowledge to fix things. Nothing is so straightforward when working with DNA, your ability to bugfix is only as good as your ability to read the code and reading DNA is a difficult and time-consuming process. Not only that, remember how I said above that we don’t know the exact relation between the DNA we put in and the DNA product we get out? If I’m trying to make a novel protein using novel DNA and it doesn’t get made, what went wrong? I can’t step through the code on this one because there’s no way to read out the activity of every RNA polymerase, every ribosome, or every post-transcriptional enzyme in the cell. I can make hypotheses and do experiments to try to guess at what is going on, but I can’t bugfix by stepping through the code, even using Green Fluorescent Protein as a print(“here”) crutch is difficult and time consuming. Even if I try to bugfix, the time lag between making a change and seeing the results can be weeks, months, or years depending on what system I’m working in, a far cry from how long it takes to compile code! A DNA-based R&D pipeline just doesn’t have the speed necessary to scale the way a coding house does, once you’ve got a program working the cost of sharing it is basically zero and the cost of starting a new project isn’t that great. That speed isn’t’ available to DNA companies yet.

This was a lot of words not just on 10x Genomics but on DNA-based companies in general. The pandemic-era highs may never be seen again for many of these companies, much like how some companies never again saw the highs of the Dotcom bubble. I think it’s important for investors to take a level-headed approach to DNA-based companies and not get caught up in the sci-fi hype. Anyone can sell you an idea, it takes a lot more work to make a product.

Quick post: naysayers aren’t always wrong

There was recently a nuclear fusion “breakthrough” which brought the naysayers out of the woodwork. The breakthrough claimed that scientists had used fusion to generate more energy than was put in. This claim, however, discounted the energy cost of the lasers used to achieve the fusion, which is like saying your company is profitable is you ignore all the salaries. Not only that, this breakthrough isn’t even on the way to creating a self-sustaining fusion reaction, it can not create a self-sustaining reaction due to the need to add and target new material in between each laser pulse. This “breakthrough” is seeming more and more like a nothingburger, and the naysayers have come out to say nay on it.

This has led to the usual backlash from the yaysayers: “they said at airplanes and steamships would never work! You’re ignorant if you don’t believe fusion won’t work!” It’s true that naysayers often laugh and disparage the geniuses of the age, they laughed at the Wright Brothers, they laughed at Edison, but remember they also laughed at Bozo the clown. Yaysayers don’t ever seem to acquiesce to the numerous promised technologies that never really worked, only focusing on those that did work and claiming a direct connection to the current one. So I thought I’d illuminate some prior failures.

Flying cars: everyone knows that the promise of flying cars never panned out despite much public mindshare and media hype. You may counter that “flying cars aren’t impossible, trying to make them is just expensive, difficult, and unnecessary” to which I say “perhaps so is fusion.” The possibility of making a toy-flying car which would never be road-legal is akin to using 300 megajoules to get 3 megajoules out of a fusion pellet, and claiming you have a breakthrough. Doable yes, but it doesn’t prove the endeavor to be doable at scale.

Antigravity elevators. Albert Einstein made several attempts at unifying the (then known) forces of the Universe together. When he started, physicists only knew about electromagnetism and gravity, but it was very enticing that these forces act so similarly in that they have infinite range and their power falls off with the square of the distance. Einstein and others theorized that there was some way to change electricity into gravity and vice versa, and charlatans/”inventors” jumped on the idea. One theory was an antigravity elevator which, by transporting passengers up and down through gravity waves instead of a moving cab, would be much more efficient and perhaps easier to maintain. Of course this idea never came to pass, not least because theories on the unification of gravity with electromagnetism were still missing half the puzzle: the strong and weak nuclear forces.

And here’s a great one: Supersonic flight transport aircraft. Now this might seem a weird one, Concorde showed it isn’t impossible, but as I’ve discussed before history has shown it to be clearly uneconomical when compared to its competitors. An idea doesn’t have to be impossible to get tossed aside, merely uneconomical.

I feel like people don’t realize how many seemingly great ideas have come and failed because they just aren’t economical even if they aren’t impossible. Fusion could well be one of those ideas, sure it works in physics but in economics who’s to say fission and renewables aren’t just objectively better? We’re still decades off even a working test reactor, and the one being planned is already about 4x over budget. Private companies have claimed they’ll come in and disrupt the industry but we had the same claims about a lot of failed projects over the years, who’s to say fusion will be any different? I know that fusion power as a scientific concept is perfectly sound, but as an engineering challenge or a profitable industry I remain skeptical.

Invest in what you know? How much do I need to know?

I’m a biochemical scientist. I’ve published papers. I’ve got degrees. As an investor, I’ve often been given the advice (whether from friends or randos on the internet) that to “invest in what you know” is the safest kind of investment. For me personally though, I’ve avoided investing in any particular biotech or med-tech companies outside of passive ETFs, because I feel like while I know a lot about biochemistry in general I don’t know enough in specific to have any kind of advantage in those areas. I know about Alzheimer’s disease, but I don’t know much about pharmacology so how would I discriminate between two Alzheimer’s drug companies I wanted to invest in? I know about CRISPR/Cas, but I don’t know enough about its delivery system in humans to feel confident that I could pick the winners in today’s more crowded CRISPR field. There are a lot of areas of biology that I feel I have a little knowledge, but not enough to give me an edge.

Maybe there’s a Dunning-Kruger effect here though, because while I can’t explain what cloud computing is besides “it’s like renting another person’s computer,” I have thrown a bunch of money into Microsoft and been happily watching it grow. I like my Microsoft products and my office suite, so I feel good enough about them that I feel they’re doing alright. Yet I clearly know a hell of a lot less about Microsoft than I do any of the biotech companies of the world, so why do I feel so confident investing here?

I don’t know, it’s hard to psycho-analyze myself, but am I making all the wrong moves? Should I focus on investing in biotech companies, confident that my background would give me an edge in picking the winners and avoiding the losers? For now, ETFs for me I guess, but I’ll keep blogging about them since they’re fun.

The nuclear fusion breakthrough that wasn’t

There was recently a nuclear fusion “breakthrough” which I just had to check out. I was disappointed to learn that this wasn’t a breakthrough at all, but a clever bit of marketing dressing up a modest scientific experiment. To explain what happened, a laboratory used around 300 megajoules of energy to create a 2 megajoule laser pulse. That pulse then hit a pellet of material, releasing 3 megajoules of energy as the pellet underwent nuclear fusion. The holy grail of fusion is a self-sustaining reaction, one necessity of such a reaction is that more energy must be released than is put in, and this experiment was hails as doing just that since the 3 megajoules of released energy is more than the 2 megajoule laser pulse. Yet that isn’t actually true because 300 megajoules went into creating that laser pulse, this is like saying a company is profitable if you ignore all salary costs. At the end of the day we want to develop a fusion reaction such that energy out > energy in, and this reaction simply did not do that.

I know why they tried to spin it this way, it’s a longstanding trick of pulsed-laser experiments to report only the amount of energy delivered by the laser, ignoring the amount of energy it takes to create that laser pulse. It makes your reactions seem a lot more efficient and feasible than they really are. But this kind of lying does the entire industry a disservice because it’s just more evidence on the pile of fusion-boosters overpromising and underdelivering. Reading this news you’d mistakenly believe we are now on the precipice of economical and available fusion power when in actuality we’re about as far as we’ve always been.

Beam Therapeutics: what’s so special about prime editing?

Beam Therapeutics is another biotech company often mentioned in the same vein as Ginkgo Bioworks, Amyris, and Twist Bioscience, and since I’ve blogged about all three of those I might as well blog about Beam. Unlike Ginkgo and Twist, Beam isn’t a shovel salesman in a gold rush, they’re actually trying to create drugs and sell them, in this case they’re trying to break into or perhaps even create the cutting edge industry of medical genetics, changing people’s genes for the better. I’ll briefly discuss the science of their technology, but I feel like the science surrounding their technology deserves the most focus.

Beam has a novel form of CRISPR/Cas gene editing called prime editing. In both normal CRISPR/Cas and prime editing, genetic information is inserted into a living organism by way of novel DNA, guide-nucleotides and a DNA cutting enzyme. The guide-nucleotides direct the information to the specific part of the genome where it is needed, the DNA cutting enzyme excises a specific segment of host DNA, and hopefully DNA repair mechanisms allow the novel DNA to be inserted in its place. These techniques always rely in part of the host’s own DNA repair mechanisms, you have to cut DNA to insert novel DNA and that cut must then be stitched back up. Most CRISPR/Cas systems create double-stranded breaks while prime editing creates just single stranded breaks, and this greatly eases the burden of the host DNA repair mechanisms allowing inserts to go in smoothly and with far less likelihood of catastrophic effects. Double stranded breaks can introduce mutations, cancers, or cause a cell to commit cell-suicide to save the rest of the body from its own mutations and cancers. Because Beam is using prime editing, their DNA editing should have less off-target effects and far less chances to go wrong.

So the upside for Beam is that they’re doing gene editing in what could be the safest, most effective way possible. The downside is that gene editing itself is still just half the battle.

When I look at a lot of gene editing companies, I quickly find all kinds of data on the safety of their edits, the amount of DNA they can insert or delete, and impressive diagrams about how their editing molecules work. I rarely see much info about delivery systems, and that’s because delivering an edit is still somewhat of an Achilles’s heel of this technology. In a lab setting you can grow any cell you want in any conditions you want, so delivering the editing machinery (the DNA, the guide-nucleotides, the enzymes) is child’s play. But actual humans are not so easy, our cells are not readily accessible and our body has a number of defense mechanisms that have evolved to keep things out and that includes gene editors. To give you an idea of what these defenses are like, biology has its own gene editors in the form of retroviruses which insert their DNA into organisms like us in order to force our body to produce more viral progeny, a process which often kills the host. Retroviruses package their edit machinery in a protein capsid which sometimes sits inside a lipid (aka fatty) envelope, and so the human body has a lot of tools to recognize foreign capsids and envelopes and destroy them on sight. These same processes can be used to recognize and destroy a lot of the delivery systems that could otherwise be harnessed for gene editing.

Some companies side-step delivery entirely, if it’s hard to bring gene editing to cells why not just bring the cells to gene editing. This was the approach Vertex Pharmaceuticals used in its sickle cell anemia drug, blood stems cells were extracted from patients and edited in a test tube, before being reinserted into the patients in order to grow, divide, and start producing non-sickled red blood cells. This approach works great if you’re working on blood-based illnesses, since blood cells and blood stem cells are by far the easiest to extract and reinsert into the human body. But for other illnesses you need a delivery method which, like a virus, is able to enter the organism and change its cells’ DNA from within.

So if Beam Therapeutics wants to deliver a genetic payload using their prime editing technology, they’re going to need a delivery system which obeys the following rules

  • It must be able to evade the immune system and any other systems which would degrade it before it finds its target cells
  • It must be able to be targeted towards certain cells so that it doesn’t have off target effects
  • It must be able to enter targeted cells and deliver its genetic package

So let’s look at the options.

Viruses have already been mentioned, and they can be engineered in such a way as to deliver a genetic package without causing any disease. However as mentioned they are quickly recognized and dispatched by the immune system whenever their are found, their protein shells being easy targets for our bodies’ adaptive immune system. Normal viruses get around this by reproducing enough to outcompete the immune system that is targeting them, but we don’t want to infect patients we just want to cure them, so using viruses that reproduce is off the table for gene editing.

A variety of purely lipid-based structures exist which can ferry a genetic package through the body. Our cell membranes are made of phospholipids, and phospholipids will naturally form compartments whenever they are immersed in water. Phospholipids also have the propensity to fuse with each other, allowing their internal compartments to be shared and anything inside them to move from one to the other. Packaging a gene editor inside phospholipids would be less likely to trigger the immune system, and they can be created in such a way that they target a particular cell type to deliver their genetic package. However random phospholipids can be easily degraded by the body, limiting how long they can circulate to find their target cell. Furthermore their propensity to fuse is both a blessing and a curse, allowing them to easily deliver their genetic package to targets but also making them just as likely to deliver it to any random cell they bump into instead. This means a lot of off-target delivery and the possibility for plenty of off-target effects

At the other end of the scale are nanoparticles made of metals or other compounds. Many methods exist to attach drugs to the outside of a nanoparticle and target that nanoparticle to a cell, however this in turn leaves the drug free to be interacted with and targeted by the immune system. For many drugs this is fine, but prime editing uses foreign proteins, DNA and free nucleotides and the body is downright paranoid about finding those things hanging around since that usually means the body has either a cancer or an infection. To that end, the body destroys them on site and triggers an immune response, which would severely curtain any use of nanoparticles to deliver a genetic package. Nanoparticles can also be designed hollow to allow for the prime editing machinery to fit snugly inside them, but this can lead to the machinery just falling out of the nanoparticle in transit and being destroyed anyway. You might say “well not a hollow sphere that fully surrounds the machinery so it can’t fall out?” But it does need to get out eventually if it wants to edit the cell, and if it’s encased in a solid sphere of metal it can’t do that. Enzymes to breach the metal would be cool but are impractical in this case.

Between these two extremes we have a number of structures made of lipids, proteins, polymers or metals, and they all struggle with one of these points. They can’t encase the machinery, or they can’t easily deliver the machinery, or they trigger an immune response, or they degrade easily, or they often cause off-target delivery. Delivery to the target is Step 0 of both prime editing and gene editing in general, and for the most part this step is still unsolved. I’ve visited several seminars where viral packages for delivering CRISPR/Cas systems were discussed, and while these seem some of the most promising vectors for gene editing they still have the problem of triggering the body’s immune system and being destroyed by it. The seminars I’ve watched all discussed mitigating that problem, but none could sidestep it entirely.

I do believe that Beam therapeutics has technology that works, their prime editing is clearly a thing of beauty. Beam is currently working on treatments for sickle cell anemia, as is Vertex Pharmaceutical, and as are most gene editing companies because it’s a blood-based disease that is amenable to bringing the cells to the gene editing machinery instead of having to go vice versa. But for anything where you can’t bring the cells to the editing, Beam isn’t quite master of it’s own fate because for prime editing to reach the cells of the body it will need to be delivered in some way and currently that’s an unsolved problem. Even a system that works to deliver some packages won’t necessarily work for all of them as size and immunity considerations change with the specific nature of the genetic package you’re delivering. I would also be worried about Beam’s cash burn, they are essentially pre-revenue and will need to do a lot of research before any of their drugs get to market or can be sold to a bigger player. I think they can survive for a long while by selling stock since their price has held up a lot better than other biotechs I’ve blogged about, but that’s good for them and not for a shareholder. As long as interest rates keep going up, I’ll treat pre-revenue companies with a wary eye.

Why do we still not know what causes Alzheimer’s disease?

Between 1901 and 1906, Alois Alzheimer began collecting data on the disease that would eventually bear his name. A patient with memory deficiency was autopsied after her death and her brain was found to contain amyloid plaques and neurofibrillary tangles. Around a half century prior in 1861, Guillaume-Benjamin-Amand Duchenne had described a disease that would bear his name, a form of muscular dystrophy, and like Alzheimer he had patient samples for study. In the next century and more both diseases would be studied and reported on, Duschenne Muscular Dystrophy was eventually linked to a single protein called dystrophin, and a number of FDA-approved treatments exist which target dystrophin and improve patient outcomes. Alzheimer’s disease was also linked to a protein, the amyloid plaques found by Alois contained a protein called amyloid beta. But while both diseases seem to have known causes, treatments for Alzheimer’s disease remain ineffective. What’s more, there is a growing body of evidence that the amyloid beta hypothesis for Alzheimer’s disease is on shaky ground. How is it that more than a century of study has not allowed us to even understand Alzheimer’s disease?

First, it must be said that the amyloid beta (Aβ) hypothesis for Alzheimer’s Disease (AD) didn’t come out of nowhere. Not only were the amyloid plaques found in Alzheimer’s patients coming from Aβ, but genetic evidence showed that the mutations associated with AD all seemed to affect the Aβ pathway. If the diagnostic criteria for AD included Aβ, and genetic evidence supported a role for Aβ, it seemed Aβ must surely be the cause of the disease. And further biochemical evidence supported a role for Aβ, for example when Aβ was shown to cause neuronal cell death in cultured nerve cells. The Aβ hypothesis even connects well with other diseases, Aβ acts as an aggregating prion and aggregating prions are known to cause other neurodegenerative diseases such as Creutzfeldt-Jakob Disease and Kuru. Note that some biochemists say a protein is only a prion if it comes from the prion gene of the human body, but like champagne this definition is expanding. So the Aβ hypothesis isn’t a hypothesis without support, it has strong biochemical evidence at the genomic and proteomic level, and fits in well with other brain diseases. It can certainly be said that Aβ proponents have ignored or downplayed evidence against the Aβ hypothesis, but that behavior is common in all disciplines. Science advances one funeral at a time.

Second, it should be recognized that AD is a difficult disease to study involving a difficult organ to study. AD affects memory and behavior by affecting the brain, those are processes and an organ that are still very opaque to us in general let alone in the context of AD. So AD is a disease we don’t understand affecting processes we don’t understand in an organ we don’t understand. Maybe we should feel grateful we even have drug candidates to begin with?

To bring this back to my own work, let me give you an example of the very small problem I am working on and the difficulties I am facing in getting data. We have a theory that there are different subtypes of AD. There is the rapid-onset (r-AD) subtype and the slow-onset or traditional (t-AD) subtype. We believe that this difference may be structural in nature, that the proteins causing r-AD and t-AD are the same but that they have different shapes. To this end, I am studying the structural variations of sarkosyl-insoluble proteins from AD patients.

OK what does that mean? I start by requesting patient samples from deceased AD patients matching either the r-AD or t-AD subtype. This is difficult because not everyone really agrees on the diagnostic criteria of these two subtypes (already we have problems!). Then once I have a patient sample, I perform a sarkosyl extraction. Sarkosyl is just a detergent like the one you wash your clothes with. A detergent can dissolve some things (like the dirt on your clothes) while not dissolving other things (like the pigments coloring your clothes). Previous studies have shown that the proteins causing AD are sarkosyl insoluble, so just like how laundry detergent will wash away dirt while leaving behind pigments, I can use sarkosyl to wash away non-AD proteins and keep the AD-causing proteins. These sarkosyl insoluble proteins include Aβ, but also include things like Tau and alpha-synuclein which some people hypothesize are the true cause of AD. The sarkosyl extraction is difficult, and I seem to fail at it as often as I succeed, am I just bad at my job or is this all really really hard? I hope it’s the latter but you never know. Then, once I’ve extracted the material I need from the patient’s brain, I use a variety of techniques to try to test our theory about AD. I can see if the extracts from r-AD and t-AD brains have different affects on neuronal organoids (artificial culture of cells that resembles an organ, in this case a brain), I can image the extracts with electron microscopy, I can take structural measurements with NMR, and so far all the data is frustratingly vague. I haven’t been at this job super long, but I can tell you I am not finding the One True Cause of Alzheimer’s disease any time soon.

And I think my struggles are fairly representative of the AD-researching community at large, or at least the ones I’ve talked to. It’s a disease that can only be studied biochemically post-mortem, the samples you get are both very limited and highly variable, it’s hard to relate the biochemistry back to the behavior and memory because we don’t have very good theories about that stuff to begin with, and we’re trying to use all the latest and greatest techniques to study this but we’re still struggling to get strong evidence to support our theories. After a century and more of study, we still don’t seem to be anywhere close to curing Alzheimer’s, we can’t really treat it, and we barely understand it. It can be frustrating and difficult work