1 00:00:00,000 --> 00:00:08,700 Welcome back to Voices of Tomorrow, the podcast where we explore the cutting edge of science, technology, and artificial intelligence. 2 00:00:08,700 --> 00:00:14,800 In recent years, AI has been nothing short of revolutionary, reshaping entire fields of study. 3 00:00:14,800 --> 00:00:23,700 In fact, AI's contributions to science have been so impactful that recently, two Nobel Prizes, one in physics and another in chemistry, 4 00:00:23,700 --> 00:00:27,800 have been awarded to researchers leveraging AI in their discoveries. 5 00:00:27,800 --> 00:00:35,000 We've discussed the profound advancements in AI-enabled protein prediction and design that led to these Nobel wins. 6 00:00:35,000 --> 00:00:42,600 But today, we dive into an area that, until recently, has remained hidden from our view, the RNA virus sphere. 7 00:00:42,600 --> 00:00:52,800 This is the world of RNA viruses, an incredibly diverse and fast-evolving group of viruses that, despite their ubiquity, have remained largely unexplored. 8 00:00:52,800 --> 00:00:58,700 Why? Because traditional methods have struggled to keep pace with their rapid evolution and diversity. 9 00:00:58,700 --> 00:01:04,000 But now, thanks to artificial intelligence, we are witnessing a monumental shift. 10 00:01:04,000 --> 00:01:13,300 AI is not only accelerating the discovery of RNA viruses but also revealing entirely new species and viral ecosystems we never knew existed. 11 00:01:13,300 --> 00:01:18,600 Today's episode will be focused on a groundbreaking study recently published in Cell, 12 00:01:18,600 --> 00:01:24,400 which used AI to document the vast, hidden diversity of RNA viruses in the environment. 13 00:01:24,400 --> 00:01:29,000 To understand the significance of this study, let's first set the stage. 14 00:01:29,000 --> 00:01:38,600 RNA viruses are everywhere. They infect nearly every form of life, plants, animals, fungi, bacteria and even other viruses. 15 00:01:38,600 --> 00:01:45,900 Some RNA viruses are well known to us, like those that cause influenza, HIV or COVID-19, 16 00:01:45,900 --> 00:01:51,200 but the vast majority of RNA viruses remain unknown and uncharacterized. 17 00:01:51,200 --> 00:01:56,100 Scientists refer to this uncharted territory as the RNA dark matter. 18 00:01:56,100 --> 00:02:02,500 So why have these viruses been so elusive? It's because RNA viruses are masters of evolution. 19 00:02:02,500 --> 00:02:11,000 They mutate and evolve much faster than their DNA counterparts, making it difficult for traditional tools to recognize and categorize them. 20 00:02:11,000 --> 00:02:19,400 Moreover, the key protein used to identify RNA viruses, RNA-dependent RNA polymerase, or RDRP, 21 00:02:19,400 --> 00:02:26,100 often evolves so rapidly that it becomes almost unrecognizable to standard sequence-based search tools. 22 00:02:26,100 --> 00:02:33,300 Without a reliable way to detect this polymerase, vast sections of the RNA virus sphere remain undiscovered. 23 00:02:33,300 --> 00:02:36,900 Now, imagine the challenge this presents to researchers. 24 00:02:36,900 --> 00:02:43,600 It's like trying to solve a jigsaw puzzle where many of the pieces are not just missing but constantly changing shape. 25 00:02:43,600 --> 00:02:46,900 This is where artificial intelligence comes into play. 26 00:02:46,900 --> 00:02:52,100 AI offers a way to make sense of this puzzle, even as the pieces shift before our eyes. 27 00:02:52,100 --> 00:03:00,100 Enter LucaProt, the AI-powered tool at the heart of this study, which is revolutionizing the way we explore the RNA virus sphere. 28 00:03:00,100 --> 00:03:06,100 LucaProt is built using deep learning, specifically leveraging the transformer architecture, 29 00:03:06,100 --> 00:03:11,600 the same architecture behind language models like GPT-3 and GPT-4. 30 00:03:11,600 --> 00:03:20,500 But instead of generating text, LucaProt is tasked with finding hidden RNA viruses in massive datasets of metagenomic sequences. 31 00:03:20,500 --> 00:03:24,700 The researchers behind LucaProt, as detailed in the cell paper, 32 00:03:24,700 --> 00:03:29,000 aim to tackle one of the most fundamental problems in viral discovery, 33 00:03:29,000 --> 00:03:34,900 identifying highly divergent RDRP sequences in vast amounts of metagenomic data. 34 00:03:34,900 --> 00:03:41,200 Traditional tools had been limited by their reliance on close-sequence similarity to known viruses. 35 00:03:41,200 --> 00:03:46,100 In contrast, LucaProt doesn't just match sequences to known examples. 36 00:03:46,100 --> 00:03:53,000 It uses AI to predict the structure and function of highly divergent sequences that have no close-known relatives. 37 00:03:53,000 --> 00:03:58,200 LucaProt integrates sequence data with structural predictions from ESMfold, 38 00:03:58,200 --> 00:04:04,400 a powerful AI tool developed by Meta that predicts the three-dimensional structures of proteins. 39 00:04:04,400 --> 00:04:07,900 By combining both the sequence and structural information, 40 00:04:07,900 --> 00:04:12,300 LucaProt is able to detect viruses that traditional models would miss, 41 00:04:12,300 --> 00:04:17,600 even when their RDRP sequences have diverged dramatically from known viruses. 42 00:04:17,600 --> 00:04:23,700 This approach allowed LucaProt to scan through 10,487 metatranscriptomes, 43 00:04:23,700 --> 00:04:29,300 datasets representing genetic material collected from diverse ecosystems around the world. 44 00:04:29,300 --> 00:04:34,200 In total, LucaProt analyzed 51 terabytes of sequencing data, 45 00:04:34,200 --> 00:04:44,900 ultimately identifying a staggering 161,979 putative RNA virus species and 180 RNA virus supergroups. 46 00:04:44,900 --> 00:04:52,700 To put this in perspective, this discovery represents one of the largest expansions of the known RNA virus sphere in history. 47 00:04:52,700 --> 00:04:56,400 Now, let's break down how LucaProt actually works. 48 00:04:56,400 --> 00:05:02,200 At its core, LucaProt uses a deep learning model based on the transformer architecture. 49 00:05:02,200 --> 00:05:08,100 What makes the transformer architecture so powerful is its ability to process sequences of varying lengths 50 00:05:08,100 --> 00:05:12,600 while capturing both local and long-range dependencies within the data. 51 00:05:12,600 --> 00:05:16,100 This makes it ideal for biological sequence analysis, 52 00:05:16,100 --> 00:05:21,700 where relationships between distant parts of a sequence can be crucial for understanding function. 53 00:05:21,700 --> 00:05:25,300 LucaProt doesn't stop at analyzing the sequence alone. 54 00:05:25,300 --> 00:05:31,200 It integrates the predicted 3D structure of proteins, using the AI model ESMfold. 55 00:05:31,200 --> 00:05:35,900 ESMfold predicts how a protein will fold into its three-dimensional shape, 56 00:05:35,900 --> 00:05:40,700 which is critical for understanding how that protein functions in the real world. 57 00:05:40,700 --> 00:05:46,200 This structural prediction is key because even if two protein sequences are highly divergent, 58 00:05:46,200 --> 00:05:50,600 their three-dimensional structures might still perform similar functions. 59 00:05:50,600 --> 00:05:57,700 For example, let's say you have two RNA viruses whose RDRP sequences look almost nothing alike. 60 00:05:57,700 --> 00:06:01,400 Traditional models would miss the connection, but LucaProt, 61 00:06:01,400 --> 00:06:04,900 thanks to its deep learning model and structural predictions, 62 00:06:04,900 --> 00:06:11,200 can recognize that both sequences fold into similar structures and therefore likely perform the same function. 63 00:06:11,200 --> 00:06:16,500 This allows LucaProt to identify highly divergent viruses that have evolved to the point 64 00:06:16,500 --> 00:06:20,900 where their sequences are barely recognizable compared to known examples. 65 00:06:20,900 --> 00:06:27,200 In fact, LucaProt was able to identify 70,000 novel viruses in just this way. 66 00:06:27,200 --> 00:06:30,800 Many of these viruses were found in extreme environments, 67 00:06:30,800 --> 00:06:34,800 such as salt lakes, hot springs, and hydrothermal vents, 68 00:06:34,800 --> 00:06:39,500 places where viral life has adapted to thrive under extreme conditions. 69 00:06:39,500 --> 00:06:44,200 Some of the viruses LucaProt discovered were unlike anything seen before, 70 00:06:44,200 --> 00:06:49,700 with genomes that stretched up to 47,250 nucleotides in length, 71 00:06:49,700 --> 00:06:53,600 far exceeding the typical size of known RNA viruses. 72 00:06:53,600 --> 00:06:57,700 To give you a better sense of the power of LucaProt's eye-driven approach, 73 00:06:57,700 --> 00:07:01,800 imagine trying to identify distant relatives at a family reunion. 74 00:07:01,800 --> 00:07:07,000 Traditional viral discovery methods would be like searching for familiar faces in the crowd. 75 00:07:07,000 --> 00:07:12,400 If someone doesn't closely resemble a family member you already know, you'd miss them completely. 76 00:07:12,400 --> 00:07:18,800 LucaProt, however, is like having a tool that can identify even the most distantly related family members 77 00:07:18,800 --> 00:07:22,200 by looking at more than just surface-level features. 78 00:07:22,200 --> 00:07:25,100 It can find commonalities at a deeper level, 79 00:07:25,100 --> 00:07:28,600 whether they share a similar bone structure or genetic markers, 80 00:07:28,600 --> 00:07:31,500 even if their appearance has changed dramatically. 81 00:07:31,500 --> 00:07:34,700 This ability to detect hidden patterns and make connections 82 00:07:34,700 --> 00:07:40,400 that humans and traditional tools might miss is exactly what makes LucaProt so revolutionary. 83 00:07:40,400 --> 00:07:44,000 It's a reminder of the power of AI not just to analyze data, 84 00:07:44,000 --> 00:07:48,100 but to discover new realms of knowledge that were previously inaccessible. 85 00:07:48,100 --> 00:07:53,400 LucaProt's discoveries don't just expand our understanding of the RNA virus sphere. 86 00:07:53,400 --> 00:07:58,100 They point to a future where AI is at the forefront of biological discovery. 87 00:07:58,100 --> 00:08:00,700 The implications of this work are profound. 88 00:08:00,700 --> 00:08:05,200 We're now able to explore viral diversity at a scale never before possible, 89 00:08:05,200 --> 00:08:11,800 opening up new avenues for research into viral ecology, evolution, and host-virus interactions. 90 00:08:11,800 --> 00:08:16,200 One of the most exciting applications of this research is in understanding the role 91 00:08:16,200 --> 00:08:19,500 these newly discovered viruses play in ecosystems. 92 00:08:19,500 --> 00:08:25,400 Are they infecting plants, animals, or even microorganisms we haven't yet identified? 93 00:08:25,400 --> 00:08:28,300 How do they influence the environments they inhabit? 94 00:08:28,300 --> 00:08:32,500 These are questions that LucaProt's discoveries are just beginning to address. 95 00:08:32,500 --> 00:08:36,000 And as AI models like LucaProt continue to improve, 96 00:08:36,000 --> 00:08:41,800 we will gain more detailed insights into the complex web of life shaped by RNA viruses. 97 00:08:41,800 --> 00:08:45,800 However, this work also highlights a number of open questions. 98 00:08:45,800 --> 00:08:50,900 For instance, while LucaProt identified tens of thousands of novel viruses, 99 00:08:50,900 --> 00:08:56,200 there remains a significant gap in our understanding of the hosts these viruses infect. 100 00:08:56,200 --> 00:09:02,800 No clear RNA virus has been shown to infect archaea, for example, leaving us with the question, 101 00:09:02,800 --> 00:09:07,500 are there RNA viruses infecting these organisms, or are they immune? 102 00:09:07,500 --> 00:09:11,500 These are the kinds of questions that future research will need to answer. 103 00:09:11,500 --> 00:09:16,500 While LucaProt has made incredible strides, we are still only scratching the surface 104 00:09:16,500 --> 00:09:19,500 of what AI can uncover in the biological world. 105 00:09:19,500 --> 00:09:25,800 Let's take a moment to step back and see where this work fits within the larger context of eye-driven discoveries. 106 00:09:25,800 --> 00:09:29,800 Earlier this year, we celebrated AI's contribution to science 107 00:09:29,800 --> 00:09:33,700 when the Nobel Prize in Chemistry was awarded for AlphaFold, 108 00:09:33,700 --> 00:09:39,000 a deep learning model that predicts the structure of proteins with unprecedented accuracy. 109 00:09:39,000 --> 00:09:42,800 LucaProt represents the next phase of this AI evolution. 110 00:09:42,800 --> 00:09:45,700 While AlphaFold focuses on known proteins, 111 00:09:45,700 --> 00:09:51,800 LucaProt pushes the boundaries even further by discovering completely unknown RNA viruses, 112 00:09:51,800 --> 00:09:54,500 many of which have never been seen before. 113 00:09:54,500 --> 00:09:58,200 In a way, LucaProt builds on AlphaFold's legacy, 114 00:09:58,200 --> 00:10:03,700 using AI to explore the hidden corners of biology that have long eluded scientists. 115 00:10:03,700 --> 00:10:09,800 Both LucaProt and AlphaFold highlight the incredible power of AI not just to solve existing problems 116 00:10:09,800 --> 00:10:14,500 but to uncover entire realms of life that were previously beyond our reach. 117 00:10:14,500 --> 00:10:17,500 AI is no longer just an analytical tool, 118 00:10:17,500 --> 00:10:21,400 it's becoming an indispensable partner in biological discovery. 119 00:10:21,400 --> 00:10:28,300 The RNA viruses uncovered by LucaProt are not simply an addition to our database of known life forms. 120 00:10:28,300 --> 00:10:31,200 They represent a new frontier in virology, 121 00:10:31,200 --> 00:10:36,900 one that could reshape how we think about viral evolution, ecology, and even human health. 122 00:10:36,900 --> 00:10:38,400 So, what's next? 123 00:10:38,400 --> 00:10:44,100 LucaProt's discovery of tens of thousands of new RNA viruses is only the beginning. 124 00:10:44,100 --> 00:10:50,500 AI is poised to play an even bigger role in biology as these tools become more refined and more powerful. 125 00:10:50,500 --> 00:10:54,300 With AI's help, we're not just uncovering hidden viruses, 126 00:10:54,300 --> 00:10:58,900 we're gaining new insights into how ecosystems function, how life evolved, 127 00:10:58,900 --> 00:11:03,600 and even how viruses might be harnessed for future biomedical applications. 128 00:11:03,600 --> 00:11:10,800 The sheer scale and speed at which AI can process and analyze biological data is unlike anything we've seen before. 129 00:11:10,800 --> 00:11:14,000 As models like LucaProt continue to improve, 130 00:11:14,000 --> 00:11:18,400 we will unlock even more secrets hidden in the biological, dark matter, 131 00:11:18,400 --> 00:11:21,400 transforming our understanding of life on Earth. 132 00:11:21,400 --> 00:11:25,000 Thank you for tuning into this episode of Voices of Tomorrow, 133 00:11:25,000 --> 00:11:32,400 where we explored the vast, hidden virus sphere and how AI is unlocking secrets that have eluded humanity for centuries. 134 00:11:32,400 --> 00:11:37,300 As we've seen, the boundaries of science are expanding faster than ever before, 135 00:11:37,300 --> 00:11:41,900 and with the help of AI, we are rewriting the very code of life itself. 136 00:11:41,900 --> 00:11:43,700 But this is just the beginning. 137 00:11:43,700 --> 00:11:50,900 The discoveries we've discussed today are a glimpse into a future where AI will lead us to answers we didn't even know to ask. 138 00:11:50,900 --> 00:11:59,300 So, whether you're a researcher on the front lines of innovation, or someone fascinated by the potential of technology to transform our world, 139 00:11:59,300 --> 00:12:01,200 I invite you to stay connected. 140 00:12:01,200 --> 00:12:07,900 Be sure to subscribe, share your thoughts, and let us know what excites you most about the future of AI. 141 00:12:07,900 --> 00:12:15,800 In the episodes to come, we'll continue our deep dive into the cutting-edge intersections of technology, biology, and beyond. 142 00:12:15,800 --> 00:12:19,300 Will we explore the mysterious depths of quantum computing? 143 00:12:19,300 --> 00:12:23,600 Or perhaps investigate the next wave of breakthroughs in eye-driven healthcare? 144 00:12:23,600 --> 00:12:26,500 Whatever it may be, I can promise you this. 145 00:12:26,500 --> 00:12:33,200 The discoveries we will uncover together will challenge what we thought we knew and inspire new visions of what's possible. 146 00:12:33,200 --> 00:12:49,900 So, join us again next time on Voices of Tomorrow, where today's voices of innovation are shaping the future, one breakthrough at a time.