AI & Computation

Instead of hitting a tumour as hard as possible, adaptive therapy gives just enough drug to keep it in check and stops when it shrinks, so drug-sensitive cells survive to compete with resistant ones. A pilot trial in prostate cancer roughly doubled the time to progression on abiraterone.

Instead of equations for average behaviour, agent-based models simulate every cell as an individual with rules for dividing, moving, dying and signalling, producing virtual tumours in which immune attack, drug delivery and evolution can be watched and tested.

Software that draws organs and tumours on scans automatically, saving hours per patient and making daily plan adaptation practical.

AI compute platforms are the GPUs, model libraries, and cloud services that pathology, radiology, and drug-design AI run on.

Software that reads scans alongside radiologists, catching cancers earlier and predicting who is at risk.

Real-time software that highlights polyps on the colonoscopy screen, helping doctors find more of the growths that could become bowel cancer.

Software that reads digitised biopsy slides to detect cancer, grade it, and score biomarkers such as HER2, PD-L1 and Ki-67 more consistently than the eye alone.

Software, increasingly LLM-based, that reads a patient's record and finds trials or guideline options they qualify for.

Software that reads screening mammograms alongside or before radiologists, catching more cancers and cutting the reading workload in large trials.

Using machine learning to pick targets, design molecules and antibodies, and predict which ADC will work.

Aidoc CARE is one radiology foundation model, pretrained on CT scans without labels, whose task-specific heads have each been FDA-cleared to flag urgent findings in emergency scans so radiologists read those first. Its oncology relevance is indirect, catching incidental masses; the regulatory evidence covers triage, not diagnostic accuracy for tumours.

Predicts the 3D shape of proteins together with DNA, RNA, small molecules and antibodies, the starting point for much modern drug design.

Reads a million letters of DNA at once and predicts how a mutation changes gene regulation, splicing and chromatin.

Scored all 71 million possible single-letter protein changes in humans as likely harmful or benign.

Models of how tumours recruit blood vessels, and of Rakesh Jain's idea that anti-angiogenic drugs at the right dose normalise rather than destroy those vessels, improving drug and oxygen delivery for a window of days.

Several randomised trials tested asking patients to report symptoms every week, with a nurse alerted when something is bad or getting worse. It reliably improves how people feel and function and cuts emergency visits. Whether it lengthens life did not hold up: two trials found a survival benefit and the largest trial, designed to test exactly that, found none.

Atlas is a pathology foundation model trained on 1.2 million slides from two of the world's largest hospitals.

BioEmu is a Microsoft generative diffusion model that predicts the range of shapes a protein moves between, not one static structure, thousands of times faster than molecular dynamics simulation. For cancer drug discovery that can reveal transient pockets, as in KRAS, that static predictors miss, but its outputs are approximate and validated mainly on small proteins.

Chemotherapy doses are usually written per square metre of body surface, a convention from 1958 that scales drug clearance between species and people; it is imprecise, and for many newer drugs flat or weight-based doses have replaced it.

Open-source structure models that match AlphaFold 3, with Boltz-2 also predicting how strongly a drug binds.

Large public collections of cancer cell lines profiled for their genomes, drug sensitivity and gene dependencies, the reference data behind much of modern target discovery.

The public systems that count every cancer diagnosis and death in a country, which tell us whether incidence and survival are improving.

Curated databases that say what each mutation means for treatment, and the expert meetings that use them to decide on therapy.

Scheduling and tracking software that makes sure each patient's cells come back to that patient, on time.

Turns a cell's gene expression into a sentence so a normal language model can reason about it; a 27-billion-parameter version proposed a cancer immunotherapy idea that was confirmed in the lab.

CellFM is an 800-million-parameter single-cell model trained on 100 million human cells.

Structure and antibody-design models from Chai Discovery, with Chai-2 reporting high zero-shot antibody hit rates.

A pathology model trained across 19 cancer types that predicts survival and mutations from slides.

Software that turns raw sequencer output into a report of which mutations matter and which drugs they point to.

Clinical trial software is the set of systems that collect trial data, randomise patients, and keep every form auditable.

Peter Nowell's 1976 idea that a tumour is an evolving population of competing clones is now measured directly by sequencing several regions or repeated blood samples, and models of branching evolution predict which clones will drive relapse.

Companies that run clinical trials for sponsors: sites, monitoring, data, and regulatory filing.

A model pretrained on 148,000 CT scans to segment organs and triage findings.

Running parts of a trial at home or locally, with telehealth, home nursing, and remote monitoring, so patients far from big centres can take part.

Magnified skin imaging and whole-body photo mapping, increasingly read by algorithms, to find melanoma early and avoid unnecessary biopsies.

Scanning microscope slides and letting software measure things a pathologist cannot see, including predictions of who will benefit from a treatment.

Using a model of what would have happened to a patient on standard treatment, so fewer people have to be randomised to it.

Patients report symptoms weekly through an app or web form, and nurses respond to alerts. Randomised trials showed this simple system improved quality of life, cut emergency visits and, in one trial, extended survival by five months.

Apps and sensors that let patients report symptoms between visits, which in trials improved survival and cut emergency visits.

Models that predict how DNA sequence controls gene activity, used to interpret non-coding cancer mutations.

ESM3 is a generative protein model that designed a working fluorescent protein far from any natural sequence.

A DNA language model trained on 9.3 trillion bases that can flag cancer-causing BRCA1 variants without being told about them.

Mathematics from population genetics shows that resistant cells almost always exist before treatment in large tumours, and that combining drugs with different resistance mutations from the start can succeed where the same drugs in sequence fail.

Cancer cells are treated as players whose success depends on what neighbouring cells do, which lets researchers predict how a tumour's mix of cell types shifts under treatment and design schedules that steer it.

Federated learning trains one AI model across hospitals by exchanging model updates, not patient data, so a pathology or radiology model learns from every site while records stay behind each firewall. Owkin, NVIDIA FLARE and the MELLODDY pharma consortium use it; governance overhead and differing data across sites are the practical obstacles.

A model trained on millions of hospital records that forecasts a patient's next diagnoses.

GEARS is a graph model predicting the effect of gene knockouts; the perturbation benchmarks around it showed how hard the problem is.

Geneformer is a transformer trained on about 30 million single cells that encodes each cell as a ranked list of its genes, so deleting a gene in silico shows which genes matter in a disease. It was the first single-cell foundation model in general use, though benchmarks find only modest gains over linear baselines on some tasks.

Uses text embeddings of gene descriptions from a general LLM to represent cells, and performs surprisingly well.

Cloud systems where hospitals and researchers store and analyse genomic data securely at petabyte scale.

Resistant cells arise by chance mutation as a tumour grows, so the chance of a cancer already containing resistant cells rises with its size. The model argued for treating early and for alternating non-cross-resistant drugs.

Tumours do not grow exponentially forever: growth slows as they enlarge, following a curve Benjamin Gompertz devised for human mortality in 1825 and Anna Kane Laird fitted to tumours in 1964. It explains why small tumours are the most chemosensitive and why doubling times lengthen.

An open 1.1-billion-parameter pathology model from a French startup, among the strongest on public benchmarks.

Hibou is a family of open pathology foundation models under a permissive licence.

A dose of chemotherapy kills a constant fraction of cancer cells, not a constant number, so each cycle removes the same proportion, which is why treatment continues after the tumour has disappeared from scans.

Mathematical oncology writes down how tumours grow, evolve, respond to treatment and interact with the immune system as equations or simulations, then uses them to design doses, schedules and trials. The models themselves are records, each with what it was fitted to and what it was used to decide.

Google's medical versions of its Gemini models, able to reason over text, images, and long records.

Adaptations of Meta's Segment Anything model that outline tumours and organs on any scan with a click.

Merlin is a model trained on 15,000 CT scans with their reports that can find and describe hundreds of findings.

From Paget's seed-and-soil idea to models that estimate when metastases were seeded from a primary and how long they lay dormant, these frameworks explain late relapse and argue for treating micrometastases early.

Midnight is a pathology model that matched the leaders while training on far fewer slides.

Reads a mammogram to estimate five-year breast cancer risk, consistently across races and devices.

A model that reads slides and clinical text together to predict who will respond to immunotherapy.

Nicheformer is a model trained on both dissociated and spatial data so it learns how a cell's neighbourhood shapes it.

Because tumours regrow fastest when small, the best way to finish them is to give the same chemotherapy doses closer together. The idea, from Larry Norton and Richard Simon, was proved in breast cancer by the CALGB 9741 trial and made two-weekly chemotherapy a standard.

DNA language models trained on thousands of genomes for variant and regulatory prediction.

NVIDIA BioNeMo is a software stack, not a model: GPU-optimised training recipes and inference services on which protein, DNA and single-cell foundation models such as Evo 2, ESM and Geneformer are trained and served. The Arc Institute, Recursion and pharmaceutical companies use it, at the price of being tied to NVIDIA hardware and tooling.

Databases built from millions of real patient records, used to see how treatments work outside trials and to run studies without new trials.

The ordering and record systems oncologists use every day, including built-in treatment pathways that steer drug choice.

The software that holds a radiotherapy patient's plan, checks that the machine settings match it before every beam is switched on, and keeps the record of every dose given. Almost every department runs one of two systems.

Pathology and radiology foundation models are AI networks pretrained without labels on over a million slides or scans (Virchow used 1.5 million), then adapted with small task heads to predict mutations, prognosis or treatment response from routine images. They power the FDA-cleared ArteraAI tools, but validation across hospitals and how regulators treat general-purpose models remain unsettled.

A digital twin is a computer model of one patient's tumour and body, updated with each scan and blood test, used to forecast how the disease will respond to each option before it is tried.

Equations that describe how a drug's concentration rises and falls in the body and how that concentration translates into effect and toxicity; the reason doses are given per square metre, why some drugs are infused over days, and how children's doses are set.

A model trained on billions of cell microscopy images to read what a drug or gene knockout does to a cell.

Owkin's open pathology models trained on TCGA and its federated hospital network.

PLUTO is PathAI's compact pathology foundation model, a vision transformer pretrained at several magnifications on 195 million tiles from 158,000 slides, so one network serves slide-level and biomarker quantification tasks at whatever resolution each needs. It runs inside PathAI's AISight product, but its weights are proprietary, so outside groups cannot benchmark or adapt it.

A score built from hundreds of common gene variants that says whether your inherited risk of a cancer is higher or lower than average, now being tested as a way to decide who is screened and how often.

Gliomas grow by both dividing and migrating through the brain, and a two-parameter equation fitted to a patient's MRI scans estimates how far invisible cells have spread, which can guide how much brain to irradiate and how fast the tumour will grow.

An open pathology model trained on 1.3 billion image tiles from a US health system, modelling whole slides at gigapixel scale.

Mechanistic computer models that join a drug's pharmacology to the biology of the tumour and the body, used by developers and regulators to pick doses, predict combinations and explain why a trial failed.

An open generalist model that answers questions about 2D and 3D scans.

Turning ordinary CT, MRI and PET scans into hundreds of measured features of shape and texture that computers relate to tumour biology and outcome.

The software that calculates exactly how radiation beams should be shaped and checks the machine delivered it.

The speed at which a molecular marker falls during treatment predicts outcome better than a single level: BCR-ABL halving time in chronic myeloid leukaemia and circulating tumour DNA slopes in solid tumours are now used to judge response within weeks.

The tools that design entirely new proteins to bind a chosen target, now used for cancer binders and antibodies.

scFoundation is a 100-million-parameter model trained on 50 million cells, from China's BioMap.

A GPT-style model for single-cell data that predicts cell types, perturbation responses, and gene networks.

Growing tumours compress themselves and their surroundings; models of this solid stress explain collapsed vessels, poor drug delivery and stiff stroma, and point to drugs that soften the tumour so treatment can get in.

Choosing treatment from a map of where each cell type sits in the tumour, not just from a list of its mutations.

Predicts how cells will respond to a drug or gene knockout, trained on over 100 million perturbed cells.

Predicts a person's six-year lung cancer risk from one low-dose CT, even when no nodule is visible.

Video visits, remote monitoring and home delivery of some cancer treatments expanded massively during COVID-19 and have stayed; they reduce travel burden, especially for rural patients, without evidence of worse outcomes.

Models trained on Tempus's paired genomic, pathology, imaging and outcome data to predict response and prognosis.

Radiotherapy works through repair, redistribution, reoxygenation and repopulation between fractions; the discovery that tumours speed up their regrowth during a course explained why gaps in treatment cost cures and led to accelerated schedules.

TITAN is a model that summarises a whole slide, not just tiles, and can write a draft pathology report.

CZI's open cross-species cell models and a reasoning model trained on them.

Curves that turn a radiation dose into a probability: how likely the tumour is to be eradicated and how likely a nearby organ is to be damaged. They underlie dose constraints, dose escalation trials and comparisons between treatment plans.

How long a tumour takes to double in volume, measured from two scans; it separates cancers from benign nodules in lung screening, sorts aggressive from indolent disease and estimates how long a tumour has been present.

Predator-prey style equations describe how immune cells hunt tumour cells, and they reproduce dormancy, escape and the delayed, sometimes explosive, responses seen with immunotherapy; they now help design combination and scheduling trials.

Two open academic pathology models: UNI reads tissue images, CONCH links images with pathology text.

Universal Cell Embedding maps any cell from any species into one shared space without retraining.

A pathology foundation model trained on millions of slides that can detect cancer and predict biomarkers from an ordinary H&E slide.

A wrist or hip device records steps, activity and sleep continuously, at home, without anyone being asked a question. It measures behaviour rather than capacity, which is the gap a corridor test leaves, and it is the one measurement of recovery that does not stop when the person leaves the hospital.

The trials finished years ago and most people being treated for cancer are still not asked their symptoms between appointments. The clearest thing that changed is a United States payment model that now requires practices to collect them.

The scanners that turn glass slides into gigapixel images, and the software that stores and serves them, without which pathology AI cannot run.