What if an artificial intelligence could understand DNA better than scientists\u2026 without ever being taught what a disease is?<\/p>\n
That\u2019s exactly what researchers claim with Evo 2<\/strong>, the largest biological AI model ever created. And its capabilities go far beyond analysis. This system doesn\u2019t just read the code of life\u2014it can write it<\/strong>.<\/p>\n The essential takeaway:<\/strong> Evo 2, a new biological AI model published in Nature<\/em>, can predict cancer-causing mutations with over 90% accuracy without ever being trained on medical data<\/strong>. Even more striking, it can generate entirely new functional DNA sequences<\/strong>, including synthetic viruses capable of killing antibiotic-resistant bacteria\u2014marking a turning point in how humanity reads and writes the code of life.<\/p>\n Evo 2 is based on the same core idea as modern language models like ChatGPT or Claude. Instead of learning human language, however, it was trained on the language of DNA<\/strong>.<\/p>\n DNA is essentially a sequence of four letters: A, T, C, and G. These letters encode everything about living organisms\u2014from physical traits to disease risks.<\/p>\n To train Evo 2, researchers compiled genetic sequences from more than 128,000 species<\/strong>, totaling around 9.3 trillion DNA letters<\/strong>. The model was trained using over 2,000 NVIDIA GPUs<\/strong> for months and ultimately reached 40 billion parameters<\/strong>.<\/p>\n But here\u2019s the twist: no labels were provided<\/strong>.<\/p>\n The AI was never told what causes disease, what a gene does, or which mutations are dangerous. It simply read raw DNA sequences\u2014and figured everything out on its own.<\/p>\n The key lies in evolution itself.<\/p>\n All the DNA sequences in the training data come from living organisms. Over billions of years, natural selection has filtered out harmful mutations while preserving functional ones.<\/p>\n This means the dataset is implicitly labeled by nature.<\/p>\n By analyzing patterns across species, Evo 2 learned to identify:<\/p>\n In short, it learned the grammar of life<\/strong> without any explicit instruction.<\/p>\n One of the most striking tests involved the BRCA1 gene<\/strong>, known for its link to breast and ovarian cancer.<\/p>\n The challenge is massive: every human carries thousands of genetic variations, and identifying which ones are dangerous is extremely difficult\u2014even for experts.<\/p>\n Evo 2, without ever being trained on medical datasets, achieved over 90% accuracy<\/strong> in distinguishing between harmless mutations and potentially deadly ones.<\/p>\n In some cases, it even matched or outperformed specialized tools designed specifically for this task.<\/p>\n And unlike traditional systems, it doesn\u2019t need to be retrained for each gene. It generalizes.<\/p>\n Evo 2 can process up to 1 million DNA letters at once<\/strong>, giving it a long-range understanding of genetic interactions.<\/p>\n This matters because in biology, a gene can be influenced by elements located hundreds of thousands of letters away.<\/p>\n In tests, the model successfully:<\/p>\n It even decoded the genome of the woolly mammoth<\/strong>, a species it had never seen before.<\/p>\n This is where things take a dramatic turn.<\/p>\n Evo 2 doesn\u2019t just analyze DNA\u2014it can generate entirely new sequences from scratch.<\/p>\n Researchers asked it to complete segments of human mitochondrial DNA. It produced hundreds of unique sequences<\/strong>, each containing all the necessary components for biological function.<\/p>\n These included:<\/p>\n Even more impressively, protein structures derived from these sequences were validated using external tools and showed correct folding behavior.<\/p>\n The model also generated:<\/p>\n This marks a major leap: AI-generated DNA that works in real biological systems<\/strong>.<\/p>\n Perhaps the most impactful application involves bacteriophages<\/strong>\u2014viruses that infect bacteria but not humans.<\/p>\n Researchers used Evo 2 to design new synthetic phages targeting antibiotic-resistant bacteria, a global health threat responsible for over a million deaths each year.<\/p>\n Out of 285 AI-generated candidates:<\/p>\n This is the first time in history that an AI has designed functional biological organisms validated in the lab<\/strong>.<\/p>\n With such power comes an unavoidable question: what prevents misuse?<\/p>\n To reduce risks, researchers excluded all dangerous pathogens from the training data. As a result, Evo 2 performs poorly when asked to generate harmful viruses\u2014it produces incoherent sequences.<\/p>\n But the situation is far from reassuring.<\/p>\n The model is fully open-source<\/strong>. Its code, data, and weights are publicly available.<\/p>\n Already, independent teams have built faster derivatives that incorporate additional biological complexity.<\/p>\n In theory, someone with enough resources could retrain a similar system with dangerous datasets.<\/p>\n No clear answer exists yet for how to manage this risk.<\/p>\n Despite the concerns, the potential benefits are enormous.<\/p>\n In medicine, Evo 2 could:<\/p>\n In agriculture, it could help design crops that are:<\/p>\n And in science, it represents a foundational shift\u2014from reading DNA to understanding and engineering it<\/strong>.<\/p>\n One of the study\u2019s co-authors compares Evo 2 to an operating system kernel: a base layer upon which countless specialized applications can be built.<\/p>\n The Human Genome Project allowed us to decode DNA.<\/p>\n Evo 2 suggests something far more profound: we are beginning to understand\u2014and rewrite\u2014the language of life itself<\/strong>.<\/p>\n Whether this leads to medical breakthroughs or new risks will depend on how this technology is handled.<\/p>\n But one thing is certain: biology has officially entered the AI era.<\/p>\n","protected":false},"excerpt":{"rendered":" What if an artificial intelligence could understand DNA better than scientists\u2026 without ever being taught what a disease is? That\u2019s exactly what researchers claim with Evo 2, the largest biological AI model ever created. And its capabilities go far beyond analysis. This system doesn\u2019t just read the code of life\u2014it can write it. The essential […]<\/p>\n","protected":false},"author":4,"featured_media":4254,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[12],"tags":[],"class_list":{"0":"post-4253","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-news"},"yoast_head":"\nAn AI That Learned Biology Without Being Taught<\/h2>\n
How Evo 2 “Understands” Life?<\/h2>\n
\n
90% Accuracy on Cancer Mutations\u2014Without Medical Training<\/h2>\n
Seeing What Other Models Miss<\/h2>\n
\n
From Reading DNA\u2026 to Writing It<\/h2>\n
\n
\n
Synthetic Viruses That Kill Antibiotic-Resistant Bacteria<\/h2>\n
\n
The Risk Nobody Can Ignore<\/h2>\n
A New Era for Medicine, Agriculture, and Biology<\/h2>\n
\n
\n
We Are No Longer Just Reading Life<\/h2>\n