A new generation of artificial intelligence is starting to read the human genome with a fluency that would have sounded like science fiction a decade ago. Instead of scanning DNA for one mutation at a time, these models can infer how stretches of code shape cells, organs and, ultimately, disease. If the early results hold up, genetic testing and drug discovery could shift from slow, hypothesis-driven guesswork to something closer to real-time simulation of biology.
That shift is not happening in isolation. It is emerging alongside massive reference datasets, new lab techniques and a broader retooling of pharmaceutical R&D around AI. Taken together, they point to a future in which the recipe for life written in DNA becomes a searchable, testable design space for medicine.
From reading DNA to predicting disease
The most striking change is that AI is no longer just classifying genetic variants, it is starting to interpret the language of DNA itself. An AI system from Google DeepMind has been trained to read long stretches of genetic code and infer how they control the molecules a cell needs to grow and function, effectively turning raw sequence into predictions about biology. By learning patterns across the genome, this model treats the four-letter alphabet of DNA as a structured language that can be parsed and, crucially, generalized to new sequences that have never been seen before, as described in reporting on an AI model that reads the recipe for life in DNA.
Other teams are pushing the same idea into the clinic. Scientists have built a New AI model that links specific genetic mutations to likely disease outcomes, allowing them to score variants by their potential to cause harm. At the Broad Institute, a system called popEVE estimates how likely each variant in a patient’s genome is to cause disease, a capability highlighted in the Highlights of a project aimed at speeding rare disease diagnosis. Together, these tools move genetic testing away from static panels and toward dynamic models that can interpret the full context of a person’s genome.
A single model as a hub for genetic testing
The most ambitious vision is to unify these capabilities into a single foundation model that can sit at the center of genetic medicine. Reporting on a system dubbed AlphaGenome describes an AI that can ingest whole genomes and infer how combinations of variants influence health, rather than treating each mutation in isolation. According to coverage asking whether such a system Could transform genetic testing and drug discovery, the model is trained directly on the instructions found in a cell, not just on curated variant lists.
That same reporting notes that AlphaGenome could help experts pinpoint genes associated with particular conditions or identify the cause of rare diseases, effectively turning it into a triage engine for complex cases. In that context, the detail that it can evaluate 43 different functional readouts of genetic activity is not just a technical flourish, it is a sign that the model is being asked to reason across many layers of biology at once. A follow up on how such a system Could reshape diagnostics underscores that the same engine could flag likely drug targets, blurring the line between testing and therapy design.
AI becomes the starting point for drug discovery
On the drug discovery side, AI is rapidly moving from a niche tool to the default starting point for new projects. One analysis argues that in 2026 identifying disease targets will begin with AI-guided models that integrate genomics, imaging and clinical data, instead of the old pattern where each dataset was analysed in isolation. That shift is framed as part of a broader moment in which, as another section of the same work notes, genomics data reaches unprecedented scale and requires new Biologica approaches rather than traditional bioinformatics code.
Industry leaders have been anticipating this pivot. Kevin Cramer, identified as Founder, CTO and CEO of Sapio Sciences, has argued that, through 2025, AI is set to transform early-stage drug discovery and let scientists interrogate complex datasets via natural language and interactive dialogs, a view captured in a profile of Kevin Cramer. Paul Hudson has similarly pointed to estimates that by 2025 artificial intelligence will drive 30% of new drug discoveries and cut costs by 25% to 50% in preclinical stages, a forecast laid out in his comments on how pharma can change by 2025, where Paul Hudson ties those gains directly to AI.
New tools to turn predictions into therapies
Reading DNA is only half the story, because the real test is whether AI can suggest interventions that restore health. One example comes from work described as New AI Tool in Diseased Cells, where a New AI system accurately identifies multiple drivers of disease and proposes combinations of drugs that can reverse disease states in cells. By searching across gene networks and compound libraries simultaneously, it turns what used to be an intractable combinatorial problem into something that can be explored computationally before any pipette is lifted.
Gene editing is undergoing a similar AI-driven upgrade. REPROCELL Launches StemEdit as a platform for Clinical Gene Editing Services and New iPSC Lines Using AI-Designed Editing System, positioning AI as the engine that designs the edits themselves. In that announcement, REPROCELL describes how the Lines Using AI and the Designed Editing System are meant to support clinical and translational applications, signalling that algorithmically planned edits are moving from theory into regulated services, as detailed in the description of how REPROCELL Launches the service.
Data, infrastructure and the road to mainstream care
For any of these models to be reliable, they need vast, well curated datasets that capture how genes behave in real tissues. Illumina has responded by introducing the Billion Cell Atlas, described as the most comprehensive single-cell view of human disease biology to date. The Atlas is designed to let researchers validate AI predictions across an unprecedented number of cells and conditions, according to Illumina’s own description of the Illumina initiative and a companion note that The Atlas will enable validation of AI models on the largest map of human disease biology so far, as further detailed in a second description of the Billion Cell Atlas.
At the same time, AI itself is evolving from static models into more agentic systems that can plan experiments and interact with lab hardware. A discussion of foundation models and agents in drug discovery describes how large biological models are being paired with robotics and lab automation so that AI-based systems can design, run and interpret experiments in a loop, a trend explored in detail in a feature on foundation models. In parallel, predictions about medicine in 2026 suggest that medical AI is moving from the Peak of Inflated Expectations to the early Slope of Enlightenment on the Gartner Hyp, with agentic AI and generative AI expected to work hand in hand with clinical expertise by late 2026, a trajectory outlined in a forward-looking assessment of the Slope of Enlightenment.
More from Morning Overview