The average adult body contains roughly 30 trillion cells, each performing around a billion chemical reactions per second. In just one minute, your anatomy undergoes so many changes, the English language doesn\u2019t have a word for the number \u2014 \u201ca hundred-billion-trillion\u201d is about the best we can do.<\/p>\n\n\n\n
In recent decades, scientists have dug ever deeper into the cell \u2014 past the nucleus and into the genome \u2014 uncovering a rich complexity within our smallest components. They have discovered that most every disorder of human health is caused not by a single straightforward factor, but rather a complex matrix of overlapping biological origins. This informational abundance presents a new kind of challenge for scientists: How do you process a gargantuan cache of biological data in an exacting, efficient, and, ultimately, superhuman manner?<\/p>\n\n\n\n
The answer: artificial intelligence. <\/p>\n\n\n\n
Scientists are now training AI models on troves of biodata and then directing them to generate novel hypotheses, run digital experiments, and make original biological discoveries around human health and function \u2014 the kind that might otherwise take decades. The 2024 Nobel Prize in Chemistry, for example, was awarded (in part) to two computer scientists at Google\u2019s DeepMind for their role in directing AI to solve the problem of protein folding<\/a>, which was previously among the most intractable enigmas in all of biology. That AI now has countless applications<\/a> for health and wellness. <\/p>\n\n\n\n Many biotech leaders now believe that the key to a healthier life lies buried within our biological data and that AI is the digital shovel that will finally dig it out. Here\u2019s how four of them are tapping into the potential of AI.<\/p>\n\n\n\n One way AI is already transforming biology is by treating proteins as if they were sentences to be decoded.<\/p>\n\n\n\n Just as natural language is made up of strings of letters and punctuation marks, proteins \u2014 the molecules that make up our cells, bones, and muscles \u2014 are made up of strings of amino acids. Biologists have learned that large language models (LLMs) \u2014 the kinds of AI models trained to spot patterns in text \u2014 can be adapted to spot patterns in these chains of amino acids.<\/p>\n\n\n\n \u201cI had always thought proteins were too complex for us to understand \u2014 certainly too complex for me to understand,\u201d said Eric Kelsic, CEO of Dyno Therapeutics<\/a>, a biotech company that uses AI to develop viral capsids<\/a> \u2014 protein shells that can be engineered to deliver gene therapies into human cells. \u201cBut with all the data, I could see there are patterns \u2014 patterns about which amino acids could be placed in certain positions on a capsid. Now, we can use AI to automate the analysis of that data and find those patterns that are difficult for us to see as humans.\u201d<\/p>\n\n\n\n “It will begin to change the way we think about our genetics.”<\/p>Eric Kelsic<\/cite><\/blockquote><\/figure>\n\n\n\n The analytic efficiency of AI means companies like Dyno could start to develop delivery systems for gene therapies to treat ultra-rare diseases \u2014 including those that have long been overlooked by the pharmaceutical industry, which focuses more attention on the diseases with the most patients. \u201cIf an AI could design a therapy for one patient and customize it to their genome, that could be done on demand, and it could be done on a massive scale,\u201d said Kelsic.\u00a0<\/p>\n\n\n\n Ultimately, AI could help enable the treatment of any kind of condition linked to genetics, from heart disease to obesity. \u201cAs we solve the challenge of being able to deliver therapeutic gene sequences into your body, it will begin to change the way we think about our genetics,\u201d said Kelsic. \u201cWe’ll start to think about our genome more around the way that we want to live or the type of person we want to become.\u201d<\/p>\n\n\n\n If AI can help decode our anatomy, it might also help to preserve it.<\/p>\n\n\n\n For more than 60 years, researchers have been testing ways to freeze human bodies<\/a> indefinitely, with the expectation that technology may one day arrive to reanimate them. But while freezing biological matter is easy \u2014 just check the meat in your freezer \u2014 freezing it without causing cellular damage is a challenge. Ice crystals have a hexagonal structure that punctures cells from within, destroying their ability to function after thawing. If we could freeze and thaw biological material without damaging it, though, the impact on healthcare would be huge \u2014 we could potentially freeze and store organs for transplants and maybe even freeze people with terminal illnesses, only reviving them after a cure is discovered. It could even be a path to living forever.<\/p>\n\n\n\n Human-led lab experiments are becoming increasingly unnecessary.<\/p>\n<\/blockquote>\n\n\n\n Biotech startup Wake Bio<\/a> is using machine learning to develop technologies that would enable the cryopreservation of whole organisms. Specifically, it\u2019s looking to develop cryoprotectants \u2014 chemical concoctions that protect cells during the freezing and thawing process. \u201cFinding cryoprotectant chemicals is the most important roadblock in getting cryonics to work,\u201d said Mark Woodward, founder and CEO of Wake Bio. \u201cAnd wonderfully, AI can help solve that.\u201d <\/p>\n\n\n\n A cryoprotectant can contain a vast number of chemicals with an essentially infinite combination of proteins, according to Woodward. AI can cut through this information thicket, identifying chemicals that might work as part of a human antifreeze and testing them in a computer model. \u201cWe’ll have experiments, the results of which will be fed into the model, which will update that model,\u201d said Woodward. \u201cIt will then be presented with a hypothetical experiment and predict whether that hypothetical experiment will succeed or not.\u201d<\/p>\n\n\n\n Woodward\u2019s in silico experiments are part of a growing trend: As AI becomes more adept at understanding not only the building blocks of our bodies, but also how they interact with each other and respond to external agents, human-led lab experiments are becoming increasingly unnecessary. If trends continue, in the future, scientists will spend less time generating hypotheses, conducting experiments, and analyzing results and more time prompting and supervising AI tools as they run digital investigations. The result might not only be an exponential leap forward in how we understand our bodies, but also how we maintain and protect them.<\/p>\n\n\n\n While Wake Bio and Dyno Therapeutics are developing AI models for very specific use cases, nonprofit research organization FutureHouse<\/a> is thinking even bigger.<\/p>\n\n\n\n \u201cThere are 20,000 genes in the human genome,\u201d said Samuel Rodriques, FutureHouse\u2019s director and CEO. \u201cYou don’t have enough time to go and read about all of them, and even if you could read about them, you wouldn’t remember in the end what you read about at the beginning. That’s what convinced me that the most important thing to do is build an AI scientist.\u201d<\/p>\n\n\n\n FutureHouse\u2019s AI models can analyze biological data, generate novel hypotheses based on the data, and then test them through digitally simulated lab experiments. This all happens in a fraction of the time and at a fraction of the cost it would take a human scientist. \u201cThis is a model for what science may look like in the future, where everything goes much faster and you have much more informed hypotheses as a result,\u201d said Rodriques. <\/p>\n\n\n\n A precisely tuned AI model could save years of work and millions of dollars in R&D.<\/p><\/blockquote><\/figure>\n\n\n\n FutureHouse is in the process of developing a whole suite of such AI science agents<\/a>. They already have models that can write code to process experimental data, help scientists dig through existing scientific literature, establish protocols for their own experiments, and even predict how different chemicals might interact with one another. It\u2019s like having an entire lab’s worth of researchers that can fit inside your pocket.<\/p>\n\n\n\n AI scientists could accelerate drug development \u2014 a process that currently requires a huge investment of time, resources, and intelligence. \u201cIn order to come up with a new drug, you need to be able to integrate information across basic biology, to cellular biology, to disease mechanics, to how you would conduct clinical trials, all the way up to whether the drug would even be insured,\u201d said Rodriques. With a precisely tuned AI model, these multidisciplinary considerations could be made in a matter of seconds, saving years of work and millions of dollars in R&D, and ultimately helping sick humans become healthier quicker. <\/p>\n\n\n\n Like Dyno Therapeutics\u2019s Kelsic, Joe Betts-LaCroix, CEO of Retro Biosciences<\/a>, identified the potential for sentence-analyzing LLMs to be able to find meaning in the chains of amino acids in proteins. \u201cIf ChatGPT can model the function and the logic behind that string of symbols, that means it could probably start modeling the logic and function behind some other types of information,\u201d he said. \u201cAnd, specifically, if it can do that for some biological datasets, then maybe we could use that to figure out how to fix the things that are broken in biology.\u201d<\/p>\n\n\n\n For Betts-LaCroix, cellular degradation \u2014 the way our cells break down as we get older \u2014 was at the top of the list of broken things to fix as the process contributes to diseases like cancer, dementia, and other age-related ailments. Reversing or preventing that degradation would require an understanding of how our cells use and recycle millions of proteins every day \u2014 a task too big for humans, but not for AI.<\/p>\n\n\n\n \u201cNobody can trace it through step-by-step exactly how it works.”<\/p>\nJoe Betts-LaCroix<\/cite><\/blockquote>\n<\/blockquote>\n\n\n\n \u201cWhat [Retro Biosciences] decided to do was to partner with OpenAI to build a bio-native model that could understand how proteins work,\u201d said Betts-LaCroix. Through this partnership, his team discovered how to reengineer Yamanaka factors<\/a> \u2014 genes that encode proteins used to reprogram cells to a state of near-nascency \u2014 to be up to 50 times more efficient. Implemented at scale, cellular reprogramming would effectively make us biologically younger and prevent nearly all the symptoms of aging, including the diseases that are most likely to kill us.<\/p>\n\n\n\n Such a breakthrough could not possibly be achieved without AI-assisted tools, or at least not achieved fast enough to make a difference in the lifespan of anyone reading this article, but there is a tradeoff: No human can point to how exact the AI models made these proteins so efficient. \u201cNobody can trace it through step-by-step exactly how it works, which has a lot of people upset or worried,” said Betts-LaCroix. <\/p>\n\n\n\n Currently, people who are wary of AI-developed drugs and therapies can still avoid them, but the tech\u2019s entanglements with medicine are only increasing. Eventually, if given the choice between an AI-developed cure for what ails them or no cure at all, even the most staunch AI skeptics may be swayed to take a chance on AI medicine \u2014 even if we humans don\u2019t fully understand how it works.<\/p>\n\n\n\nProteins as sentences<\/h3>\n\n\n\n
Cracking the cold case<\/h3>\n\n\n\n
\n
Can AI outperform scientists?<\/h3>\n\n\n\n
Reprogramming aging<\/h3>\n\n\n\n
\n
\n