Researchers at Oxford University have developed an artificial intelligence system that extracts molecular information directly from cell images, potentially accelerating cancer drug discovery and reducing reliance on costly sequencing technologies. The breakthrough, called PhenoSeq, learns the relationship between how cells look under a microscope and their underlying genetic activity, generating the same molecular insights that typically require expensive lab work.

How Does This AI System Work?

Modern drug discovery relies heavily on imaging experiments that photograph cells after exposure to different treatments. These experiments are fast and scalable, generating enormous amounts of biological data. However, understanding how cells respond at the molecular level traditionally requires specialized sequencing technologies that are significantly more expensive and time-consuming than imaging alone.

PhenoSeq bridges this gap by using artificial intelligence to predict gene-expression patterns, also called transcriptomic profiles, from cellular images without requiring additional sequencing experiments. The system learns from matched datasets containing both cell images and molecular measurements, allowing it to generate biologically meaningful molecular representations from visual data alone.

"Cell morphology and gene expression are fundamentally different measurements of the same underlying biology. Our goal was to determine whether information contained in large-scale imaging experiments could be translated into a molecular representation that is normally only accessible through costly sequencing technologies," said Dr. Tapabrata Rohan Chakraborty, a Lecturer in Computer Science at Christ Church, University of Oxford.

Dr. Tapabrata Rohan Chakraborty, Lecturer in Computer Science at Christ Church, University of Oxford

What Evidence Supports This Approach?

The research team, working with collaborators from The Alan Turing Institute and The Institute of Cancer Research in London, tested PhenoSeq using a newly released dataset containing matched cellular imaging and transcriptomic measurements across a range of chemical treatments. The results demonstrated that the AI-generated molecular profiles captured biologically meaningful information and improved the ability to distinguish between different treatments compared with imaging data alone.

This work builds on Dr. Chakraborty's earlier research called PathGen, which showed that molecular information could be generated from digital pathology images and was published in Nature Communications earlier in 2026. However, PhenoSeq represents a significant advance because it is among the first systems to generate transcriptomic representations from high-content cellular imaging specifically in the context of phenotypic drug discovery, where researchers test how cells respond to potential drug candidates.

How Could This Transform Drug Development?

The potential applications extend across multiple stages of pharmaceutical research and development. By extracting additional biological insight from existing imaging datasets without the need for extensive molecular profiling, scientists could streamline their workflows and reduce costs. The approach could support more efficient drug-screening pipelines, improve understanding of how experimental treatments work, and accelerate the search for new therapies.

The research highlights the growing potential of generative AI, a type of artificial intelligence that can create new data or representations based on patterns it learns from training data, to integrate different forms of biological data and uncover information that would otherwise remain hidden within routine laboratory experiments.

Steps to Implement AI-Driven Molecular Profiling in Drug Discovery

• Leverage Existing Imaging Data: Organizations can apply PhenoSeq and similar systems to their existing cell imaging datasets to extract molecular insights without conducting new sequencing experiments, reducing both time and cost.
• Integrate with Current Screening Pipelines: Drug discovery teams can incorporate AI-generated molecular profiles into their existing high-throughput screening workflows to improve treatment differentiation and identify promising drug candidates more efficiently.
• Combine Multiple Data Modalities: Researchers can use AI systems that integrate imaging, molecular, and other biological data types to gain a more complete understanding of how cells respond to treatments compared with any single data source alone.

The study, titled "Cell Painting Generates Single-Cell Transcriptomics via Conditional Diffusion," was accepted for presentation at the International Conference on Machine Learning (ICML), one of the world's leading conferences for machine learning research. The work was supported by the Turing-Roche strategic partnership between Roche Pharmaceuticals and The Alan Turing Institute, the United Kingdom's national institute for artificial intelligence.

Dr. Chakraborty, who serves as a Theme Lead in Frontier AI Assurance at the Alan Turing Institute, continues to advance the intersection of AI and biological research. As pharmaceutical companies and research institutions increasingly adopt AI-driven approaches to drug discovery, systems like PhenoSeq demonstrate how machine learning can reduce barriers to accessing molecular-level insights and accelerate the path from laboratory discovery to clinical application.