The same AI that keeps you scrolling TikTok might soon design your medicine. That sounds like a headline from a dystopian sci-fi novel, but it's exactly what happened last week in Boston. ByteDance — yes, the Chinese company that owns TikTok — walked into one of the world's most prestigious immunology conferences and presented a molecule that shouldn't exist. A small molecule targeting IL-17, a protein-protein interaction that pharmaceutical companies have been trying to drug for two decades. Every attempt failed. Every lab gave up. And then a TikTok company, of all things, showed up with an AI-designed compound that actually works.
The presentation came from Anew Labs, ByteDance's quietly launched drug discovery unit. The company has spent the past year assembling a team of computational biologists, medicinal chemists, and machine learning engineers in a bid to do something that sounds absurd on paper: use the same recommendation algorithms that understand your dance preferences to understand molecular binding. And against all odds, it seems to be working.
The Undruggable Target
To understand why this matters, you need to understand IL-17. It's a cytokine — a signaling protein that plays a central role in autoimmune diseases like psoriasis, rheumatoid arthritis, and inflammatory bowel disease. Blocking IL-17 has been one of pharma's holy grails because the existing treatments, mostly biologic antibodies, are expensive, require injection, and don't work for everyone.
The problem is that IL-17 is a protein-protein interaction (PPI), and PPIs are notoriously difficult to target with small molecules. The binding surfaces are flat and featureless, lacking the deep pockets that traditional drugs can nestle into. Pharmaceutical companies have spent billions trying to find a small molecule that can disrupt IL-17 signaling. Novartis tried. Pfizer tried. Genentech tried. All of them failed.
The scientific consensus was that IL-17 was "undruggable" by small molecules. The only viable approach was biologics — large antibody drugs that block the protein but must be administered by injection and cost thousands of dollars per dose. For patients, that means monthly shots, refrigeration requirements, and insurance battles. For pharma, that means a profitable but limited market.
And then ByteDance showed up with a generative AI model that designed a small molecule from scratch. In simulation, it binds to IL-17 with an affinity that rivals existing biologics. In cell assays, it disrupts signaling. In animal models, it reduces inflammation. The data is early — this is still preclinical — but the conference audience, packed with immunologists who've spent careers on this exact problem, reportedly went quiet when the results appeared on screen.
From Recommendation Algorithms to Molecular Design
The leap from TikTok to drug discovery isn't as crazy as it sounds. ByteDance's core competency isn't social media — it's pattern recognition at scale. The company's recommendation engine processes billions of user interactions per day, learning to predict what content will keep any given user engaged. That requires understanding complex, high-dimensional relationships between entities (users, videos, interests) and optimizing for specific outcomes (engagement, retention, watch time).
Molecular design, it turns out, is structurally similar. Instead of predicting which video a user will watch, you're predicting which molecular structure will bind to a protein target. Instead of optimizing for engagement time, you're optimizing for binding affinity, selectivity, and drug-like properties. The underlying mathematics — graph neural networks, attention mechanisms, generative models — is largely the same.
Anew Labs has taken this insight and built a platform that treats molecules like content recommendations. The system ingests structural data about protein targets, learns the "preferences" of the binding site, and generates molecular candidates that are predicted to "engage" the target most effectively. It then simulates how these candidates will behave in the body — solubility, metabolism, toxicity — and iterates until it finds compounds that satisfy all constraints.
The approach is radically different from traditional drug discovery, which relies on medicinal chemists manually tweaking known molecular scaffolds. Anew's AI can explore chemical space that human chemists would never consider — unusual ring structures, exotic functional groups, geometries that violate conventional wisdom but satisfy the physics of the binding site. This is how it found something that every human expert missed.
Why Big Tech Keeps Cracking Biology
ByteDance isn't the first tech giant to wander into drug discovery with surprising success. Google's DeepMind used AlphaFold to predict the 3D structure of nearly every known protein — a breakthrough that solved a 50-year-old grand challenge in biology. Meta's AI research team built ESMFold, a competing protein structure prediction system. Microsoft has partnerships with dozens of biotech companies applying cloud AI to molecular design.
The pattern is becoming clear: AI systems trained on massive datasets turn out to be remarkably good at biological prediction tasks, even when they weren't explicitly designed for them. The same transformers that process language can process protein sequences. The same diffusion models that generate images can generate molecular structures. The same reinforcement learning that masters chess can optimize molecular properties.
For traditional pharmaceutical companies, this is an existential threat. The drug discovery industry has long operated on the assumption that finding new medicines requires specialized expertise, decades of experience, and massive R&D budgets. Anew Labs suggests that a team of AI researchers with no pharmaceutical background can match or exceed that expertise in months, not decades. The implication is that drug discovery may be less like bespoke craftsmanship and more like a scalable engineering problem.
Of course, there's a long road between a conference presentation and an approved drug. The IL-17 molecule still needs to pass through years of preclinical testing, human trials, and regulatory review. Most promising drug candidates fail somewhere along that path. But the speed and cost advantages of AI-driven discovery are substantial. What used to take years and hundreds of millions of dollars can now happen in months for a fraction of the cost.
The TikTok Problem
There is, inevitably, a geopolitical dimension. ByteDance is a Chinese company operating under intense scrutiny in the United States. The company has faced accusations — some substantiated, some not — of sharing user data with the Chinese government, censoring content, and engaging in information warfare. The idea that the same company could design medicines that Americans put in their bodies is politically fraught, to say the least.
Anew Labs is structured as a separate entity, and the company has emphasized that its research is conducted independently with international collaborators. But the skepticism will be intense. If ByteDance-developed drugs ever reach the US market, they will face regulatory and political headwinds that no amount of clinical data can overcome. The FDA doesn't just evaluate safety and efficacy — it operates in a political environment, and ByteDance's brand is toxic in Washington.
The more likely path is that Anew Labs partners with Western pharmaceutical companies to develop and commercialize its discoveries. This is the model that other AI drug discovery companies have followed: find the molecule, then hand it off to a pharma giant with the resources and regulatory expertise to bring it to market. ByteDance gets licensing revenue and validation; the pharma company gets a pipeline of promising candidates; patients get new medicines. Everyone wins, at least in theory.
But even partnership models face skepticism. American pharma companies are cautious about technology transfer to Chinese entities, and the political environment is getting more hostile, not less. Anew Labs may find that its scientific breakthroughs are easier to achieve than its commercial strategy.
🔥 Our Hot Take
Here's the uncomfortable truth that pharma executives are whispering in boardrooms but won't say in public: their entire business model is being disrupted by companies that don't even consider themselves pharma companies. ByteDance didn't spend decades building medicinal chemistry expertise. They spent a decade building AI systems that understand patterns. And it turns out that's actually the relevant skill.
The drug discovery industry has always been about finding needles in haystacks — screening millions of compounds hoping one works. AI doesn't just make the search faster; it changes the nature of the search. Instead of rummaging through haystacks, AI can design needles that don't exist yet, optimized for the exact haystack you're searching. That's not an incremental improvement. That's a paradigm shift.
The ByteDance presentation should have been impossible. A TikTok company cracking an undruggable target that Pfizer and Novartis gave up on? That doesn't fit any narrative about how drug discovery works. But it fits perfectly with what we're learning about AI: the most important applications often come from unexpected places, built by teams that weren't constrained by conventional wisdom about what can't be done.
Whether Anew Labs ultimately succeeds as a business is an open question. The geopolitical barriers are real, the regulatory path is long, and most early-stage drug candidates fail. But the scientific signal is unmistakable: AI-designed molecules are arriving faster than anyone expected, and they're targeting problems that human expertise couldn't solve. The implications for patients, for pharma, and for the future of medicine are profound. The TikTok algorithm that kept you up until 2am last night might also save someone's life. That's not a future anyone predicted. But it's the future we're getting.