Artificial intelligence is fundamentally changing how scientists interpret DNA by treating the genome like a language that can be learned and predicted. Two of the world's leading AI research labs, DeepMind and IBM, are developing foundation models that scan long genetic sequences to link patterns in the code to biological behavior, from gene regulation to disease risk. This shift mirrors how large language models transformed software development, and the implications for drug discovery and personalized medicine could be transformative. What Makes AlphaGenome Different From Earlier Genomic AI Tools? DeepMind recently published AlphaGenome, a model that takes up to one million DNA base pairs as input and predicts thousands of molecular properties across diverse biological processes. In a study published in Nature, the model outperformed existing tools in 22 of 24 variant effect prediction tasks, marking what researchers describe as a fundamental shift in how scientists can interrogate the regulatory code embedded in non-coding DNA. The breakthrough lies in AlphaGenome's multimodal design. Unlike earlier genomic models that handled different types of biological data separately, AlphaGenome learns from multiple types of biological measurements simultaneously. This means it can predict effects across different genomic signals and treat them as connected rather than independent. "What I found most novel about AlphaGenome was its multimodal nature. The fact that it is trained on data from many different genomic modalities, for instance, RNA-seq, ATAC-seq and Hi-C, and predicts effects across these modalities is particularly notable," said Mark Gerstein, the Albert L Williams Professor of Biomedical Informatics at Yale University. Mark Gerstein, Albert L Williams Professor of Biomedical Informatics, Yale University Gerstein emphasized that what's truly novel is the scale at which AlphaGenome folds these relationships directly into sequence-to-function prediction. The model can "see" an unusually large window of DNA in one pass, on the order of a megabase, which is a span big enough to capture regulatory effects that can sit far from the genes they influence. Why Does Understanding Non-Coding DNA Matter So Much? The human genome contains roughly three billion base pairs, but only about two percent of them encode proteins. The remaining 98 percent orchestrates when, where and how much of each protein gets made. Small variations in this regulatory machinery can profoundly alter an organism's response to its environment or susceptibility to disease. Until recently, deciphering exactly how these sequences work at the molecular level has remained one of biology's most stubborn puzzles. Before AlphaGenome, researchers faced a difficult trade-off: scan a long region of DNA but lose fine detail, or zoom in tightly and miss the long-range signals that matter in regulation. DeepMind designed AlphaGenome to avoid that choice entirely, using an architecture that combines convolutional layers to capture short DNA motifs, transformers to share information across the entire sequence, and additional layers to translate detected patterns into predictions across multiple biological readouts. How Are IBM and DeepMind Taking Different Approaches to Genomic AI? While DeepMind is building a single system meant to read regulatory DNA as a unified code, IBM is taking a more modular approach. IBM Research has been developing its own suite of biomedical foundation models to tackle complementary challenges in drug discovery, with applications ranging from antibody design to small-molecule property prediction. "Our work on Biomedical Foundation Models takes a more practical, modular approach. We decompose complex biological questions into well-defined components and identify the mathematical and algorithmic innovations required for the specific tasks at hand," explained Michal Rosen-Zvi, Director of AI for Healthcare and Life Sciences at IBM Research. Michal Rosen-Zvi, Director of AI for Healthcare and Life Sciences, IBM Research IBM develops specialized models tailored to distinct domains, each designed to optimally capture the modalities most relevant to its domain, whether that is primary sequence, two-dimensional structure, three-dimensional conformation or mathematical representations that capture whole-genome expression at the cellular level. A key difference in IBM's approach is how it handles genetic variation. Rather than treating the genome as a single "standard" sequence, IBM explicitly incorporates population-level variation, training not only on reference sequences but also on single nucleotide polymorphisms (SNPs) and other mutable sites. This design lets the models learn evolutionary and functional signals that a single reference genome cannot capture, signals that might otherwise require training on many thousands of whole genomes. What Are the Practical Implications for Drug Discovery and Medicine? For pharmaceutical companies and biotech firms, the promise of these AI models is immense. The potential benefits include: - Faster Mutation Identification: Quicker identification of disease-causing mutations that would have taken researchers months or years to pinpoint manually. - Precise Drug Targeting: More accurate targeting of drugs to specific genetic variants, reducing side effects and improving efficacy for patient subgroups. - Computational-Guided Experiments: The ability to design experiments guided by computational predictions rather than brute-force screening, saving time and resources. - Regulatory Understanding: Better comprehension of how non-coding DNA variants affect disease risk, enabling preventive medicine strategies. DeepMind emphasized how efficiently AlphaGenome can be trained. The company said training took about four hours and used roughly half the compute budget of its earlier Enformer model, an efficiency gain that is notable given AlphaGenome's expanded scope and capability. How Does This Compare to AlphaFold's Impact on Biology? AlphaGenome arrives in the wake of AlphaFold, the protein-structure system that helped convince the world that AI could tackle parts of biology once thought too complex to model directly. But DNA is a different kind of challenge. A change in sequence does not simply alter a static structure. It can ripple through regulation, shifting when a gene turns on, how much RNA gets made, how much protein is produced and how a cell reacts to signals from its environment. Most genomics tools are built to handle that complexity in slices: one method to find protein-coding regions, another to interpret variants, another to estimate disease risk and another to support clinical decisions. AlphaGenome is designed to bring many of those steps into a single framework, rather than forcing researchers to stitch together separate models. This represents a conceptual shift similar to how large language models unified multiple natural language processing tasks. Steps to Understanding How These Models Work in Practice - Input Stage: Researchers provide a DNA sequence up to one million base pairs long to AlphaGenome, which is roughly the size needed to capture long-range regulatory effects. - Processing Stage: The model uses convolutional layers to identify short DNA motifs, transformers to share information across the entire sequence, and additional neural network layers to translate patterns into biological predictions. - Output Stage: AlphaGenome predicts gene-related features across different cell types, including signals related to transcription and aspects of RNA processing, and shows how those outputs change when the sequence is altered. - Validation Stage: Predictions are compared against experimental data from decades of publicly funded research to verify accuracy before being used to guide new experiments. AlphaGenome is trained on an enormous archive of molecular biology experiments generated over decades of research, many of them produced by publicly funded consortia. DeepMind has described using large public datasets that measure how sequence and variation relate to signals such as RNA output and transcription factor binding in human and mouse cells. By learning from these experimental patterns, the model claims to identify not only the stretches of DNA that encode genes, but also the regulatory sequences that control when genes turn on, where they turn on and how strongly. The convergence of these approaches from DeepMind and IBM signals that the genomics field is entering a new era where AI models can read DNA with unprecedented nuance and scale. As these tools mature and become more accessible to researchers worldwide, they could accelerate the pace of biological discovery and reshape how we understand disease at its genetic roots.
