Early Detection
A Single Blood Test That Hunts Twelve Cancers at Once
GRAIL's Galleri test — which analyses circulating tumour DNA in a blood draw — reached a landmark milestone in October 2024 when the PATHFINDER 2 trial reported results from nearly 36,000 participants. When added to standard USPSTF-recommended screening, Galleri detected cancers at a seven-fold higher rate, with a specificity of 99.6% (a false-positive rate of just 0.4%) and the ability to correctly identify the tissue of origin 92% of the time. The test showed 73.7% sensitivity across twelve of the cancers responsible for two-thirds of all cancer deaths in the United States — cancers including ovarian, pancreatic and oesophageal, which currently have no routine screening at all. For these "silent killers," a positive signal from a routine blood draw could be the difference between a Stage I and a Stage IV diagnosis.
GRAIL — PATHFINDER 2 Trial Results, Oct 2024AI Mammography Finds 17.6% More Breast Cancers in Real-World Study
A landmark real-world study published in The Lancet Digital Health followed 463,094 women across twelve German screening centres and 119 radiologists. When AI assisted radiologists in reading mammograms, the cancer detection rate rose from 5.7 to 6.7 per 1,000 women — a 17.6% relative increase — with no corresponding rise in recall rates or unnecessary biopsies. The positive predictive value of biopsy actually improved, from 59.2% to 64.5%. Crucially, this wasn't a controlled trial with hand-picked radiologists: it was routine clinical practice, at scale, across an entire national screening region. The AI was doing what no single radiologist can do consistently across thousands of cases — maintaining perfect attention.
The Lancet Digital Health, 2024"We are moving from a world where we find cancer by accident — or when it announces itself with symptoms — to a world where AI finds it while it is still invisible to every other tool we have." — Based on findings from the GRAIL PATHFINDER 2 Trial & Lancet Digital Health, 2024
Cracking the Undruggable
AlphaFold's Nobel Prize — and What Comes Next for Cancer
When the 2024 Nobel Prize in Chemistry was awarded to Demis Hassabis and John Jumper of Google DeepMind for AlphaFold, the oncology world paid close attention. AlphaFold 3, released in May 2024, extended the system's protein-structure predictions to DNA, RNA, and small drug-like molecules — unlocking cancer targets that had resisted every previous approach because their three-dimensional shapes were simply unknown. Alphabet's drug-discovery spin-out Isomorphic Labs followed in March 2025 with a $600 million funding round to advance AI-designed molecules into human clinical trials, specifically targeting cancers driven by hard-to-treat protein interactions. The era of drugs designed atom-by-atom on a computer and validated in silico before a single experiment is run has arrived.
Google DeepMind · Nobel Prize in Chemistry 2024BBO-10203 — An AI-Designed Drug Against a Target That Defeated Medicine for 40 Years
The Ras protein is mutated in roughly 30% of all human cancers, including the most lethal — pancreatic, colorectal and lung. For four decades, Ras was considered "undruggable": its smooth, featureless surface gave small molecules nowhere to bind. In 2024, Benzene Therapeutics entered human clinical trials with BBO-10203, a first-in-class drug that blocks Ras from activating PI3Kα — a key growth signalling pathway — using an AI-designed molecular structure that took only six years from concept to clinic, compared to the typical ten to twenty. The approach validates a new category of oncology drug design in which machine learning identifies cryptic binding sites invisible to traditional structural biology, and generative chemistry then sculpts a molecule to exploit them.
ScienceDaily · Benzene Therapeutics, 2024Precision Treatment
SCORPIO — The AI That Predicts Whether Immunotherapy Will Save Your Life
Immune checkpoint inhibitors — drugs like pembrolizumab that unleash the body's own immune system against tumours — have transformed cancer care for some patients, but for others they do nothing. Identifying in advance which group a patient belongs to has been one of oncology's hardest problems. In 2025, the NCI published results for SCORPIO, an AI tool that predicts 2.5-year survival following checkpoint inhibitor treatment with 72 to 76% accuracy across diverse patient cohorts. Remarkably, the model achieved this using inputs available at any hospital: age, sex, BMI, and standard blood test measurements — no expensive genomic sequencing required. This means the tool could function as a triage filter at the point of care, routing patients toward the therapies most likely to benefit them and away from months of ineffective treatment.
National Cancer Institute, 2025Real-Time AI Trial Matching — Closing the Gap Between Diagnosis and Precision Medicine
One of oncology's most persistent failures is that patients who could benefit from genomically matched clinical trials never learn they exist. Emory University's Winship Cancer Institute published results in 2025 from a randomised study showing that when an AI platform detected cancer progression on imaging reports and automatically flagged matching clinical trials to oncologists, appropriate enrolment rates increased significantly. The system works by reading radiology and pathology reports in real time, cross-referencing tumour genomics against open trial registries, and generating a notification within hours. It is, in effect, an always-on precision medicine navigator — ensuring that the right trial finds the right patient at exactly the right moment in their disease course.
Winship Cancer Institute, Emory University, 2025"Precision oncology used to mean sequencing a tumour and hoping a matching drug existed. Now AI is simultaneously designing the drugs, finding the patients, and predicting the outcomes — compressing what used to take careers into years." — Synthesis of findings: Isomorphic Labs, NCI SCORPIO, Winship Cancer Institute, 2024–2025
Buy me a coffee