This 👇👇👇 is one of the most exciting pieces of research my lab was involved in during the past few years — and a step change for how we study the growth and evolution of human cancers. Great work by the team!.

Hi everyone, it's time for me to join the X exodus towards bsky. I have greatly enjoyed the scientific discussion here but the backdrop has turned too dark. I'll be posting news about work and scientific musings on So long, and thanks for all the fish.

Join millions who have switched to Grok.

RT @OpenTargets: NEW: Postdoctoral Fellow with @ewanbirney. The project is based in the Birney Research group at @emblebi and working close….

RT @TimCoorens: Very happy and excited to announce that I'll be starting my own research group at @emblebi! The group will focus on lineage….

RT @nfusi: We’re excited to share our latest work with @paige_ai on (vision) foundation models of pathology. In this study, we introduce V….

If you’re German speaking and want to learn how AI is being used for cancer research, consider this podcast by @DLF to which I contributed. There are many opportunities but also some hurdles to overcome.

RT @artem_shmatko: How much does your medical history reveal about your future?. As a team effort of @MoritzGerstung and @ewanbirney's labs….

New preprint: A modified GPT architecture to model the course of health and more than 1000 diseases throughout life. Delphi-2M estimates the risk of almost any disease, samples health trajectories and offers insights into the temporal dependencies of co-morbidities. More below.

RT @jnkath: Weekend read: We are discussing the future of AI in cancer research in @Cancer_Cell . with @MoritzGe….

RT @emblebi: Dreaming of leading your own research group in computational biology & bioinformatics?. We’re recruiting two research group le….

This was a great collaboration with @ewanbirney, @udansk, Soren Brunak and many others from Denmark without whom the work would not have been possible. 🙏

As a whole the study underscores the potential benefits of systematic national health data registries for the purpose of multi-cancer risk modeling. Obviously such resources will also inform on much wider health questions beyond cancers.

The Danish registries are a phenomenal resource, but tend to lack the basic things - smoking status, alcohol consumption or weight. These data had to be extracted from free text records and were only available for a subset of individuals.

There are limitations to our analysis. Correlation is not causation. Many diagnoses can be surrogates of underlying risk factors or indicators of symptomatic cancer.

This is especially the case as new blood based tests using circulating tumour DNA are being trialled. Such tests show great promise but are unlikely to be a silver bullet. The more targeted their use the better will be the performance (fewer false positives, more true pos).

Why do these things matter?. Early cancer diagnosis is associated with better outcomes. Cancer screening, however, is very challenging and can also lead to over-diagnoses and harm. Hence understanding who is most likely to develop cancer is very important.

Interestingly the contribution to risks shift over time. At younger ages < 45yr, family history of cancers tends to be very informative. With increasing age, disease history becomes more informative as people rack up diagnoses that are indicative of elevated risks.

⚠️ These associated variables are reproducible across cohorts, but are not necessarily causal and could indicate other exposures. ⚠️

A closer look revealed that there are many factors across the categories of prior disease, family history of cancer and basic health, which are indicative of elevated cancer risks.

Remarkably for such complex models there was an excellent validation in the UK Biobank.

All 20 cancers were, to an extent, predictable with 19 showing a clear improvement over an age and sex baseline.