In silico generation of synthetic cancer genomes using generative AI
TL;DR
Imagine you have a big puzzle, but you can't see all the pieces because they're hidden for privacy reasons. This makes it hard to solve the puzzle. Scientists have found a way to create new puzzle pieces that look just like the hidden ones, so they can share them with others to help solve the puzzle faster. This means they can understand cancer better and find new ways to treat it.
Understanding how genomic alterations drive cancer is key to advancing precision oncology. To detect these alterations, accurate algorithms are used; however, due to privacy concerns, few deeply sequenced cancer genomes can be shared, limiting benchmarking and representing a major obstacle to the improvement of analytic tools. To address this, we developed OncoGAN, a generative AI model combining adversarial networks and variational autoencoders to create realistic synthetic cancer genomes. Trained on large-scale genomic datasets, OncoGAN accurately reproduces somatic mutations, copy number alterations, and structural variants across cancer types while preserving donors' privacy. The synthetic genomes reflect tumor-specific mutational signatures and positional mutation patterns. Using DeepTumour, we validated the synthetic data's fidelity, showing high concordance between generated and predicted tumors. Moreover, augmenting the training data with synthetic genomes improved DeepTumour's accuracy, underscoring OncoGAN's potential to generate shareable datasets with known ground truths for benchmarking and enhancement of cancer genome analysis tools.
- 1OncoGAN is a multimodel ensemble pipeline combining GANs, TVAEs, and random sampling that generates realistic synthetic cancer genomes for eight distinct tumor types, accurately reproducing somatic mutations, copy number alterations, and structural variants.
- 2Synthetic genomes produced by OncoGAN faithfully replicate tumor-specific mutational signatures, genomic positional mutation patterns, and driver mutation frequencies and intercorrelations observed in real PCAWG data.
- 3On average, only 0.021% of simulated mutations exactly match those in the training set, demonstrating effective donor privacy protection, making the synthetic genomes fully open access.
- 4DeepTumour achieved nearly 100% tumor-type prediction accuracy on OncoGAN-generated synthetic donors, validating the biological fidelity of the simulated genomes.
- 5Augmenting DeepTumour training data with 100 OncoGAN-simulated donors per tumor type improved overall classification accuracy from 89.26% to 90.16%, with the largest gains for underrepresented tumor types such as Lymph-MCLL (F1 score: 75% to 84%).
Single-minus gluon tree amplitudes are nonzero
Imagine tiny particles called gluons are like spinning tops. Their spin can be in one of two directions, which physicists call 'plus' or 'minus'. For decades, the rulebook seemed to say that you could never have a situation where just one gluon was spinning 'minus' and all the others were spinning 'plus' — that outcome was thought to be zero. This paper found a loophole. Under very specific, purely mathematical conditions that don't exist in our physical reality but are useful for calculations, this interaction can happen. The researchers wrote down the exact recipe for it, fixing a small but important detail in our fundamental rulebook for how the universe works.
Sub-part-per-trillion test of the Standard Model with atomic hydrogen
Scientists made an incredibly precise measurement of light emitted by hydrogen atoms that tested one of physics' most fundamental theories - the Standard Model - to an accuracy of 0.7 parts per trillion. This measurement also resolved a long-standing disagreement about the size of protons by confirming the smaller value found in previous experiments with exotic atoms.
Rock art from at least 67,800 years ago in Sulawesi
Imagine finding a spray-painted handprint on a cave wall. Over thousands of years, a thin, glassy layer of minerals, like limescale in a kettle, grew on top of it. Scientists used a high-tech laser to analyze that mineral layer. By measuring the natural radioactive decay of elements within it, they figured out the layer is about 71,600 years old. Since the handprint is underneath that layer, it must be at least that old, with the most conservative estimate being 67,800 years. This makes it one of the oldest pieces of art ever found and proves that the early humans who lived on this Indonesian island, who had to cross the ocean to get there, were creating symbolic art.
An interstellar energetic and non-aqueous pathway to peptide formation
Imagine you have a box of LEGO bricks, which are like the basic molecules of life called amino acids. To build anything, you need to snap them together. Scientists used to think you needed a puddle of liquid water to make the bricks 'click'. This experiment is like discovering you can snap the LEGOs together inside a freezer. The researchers took the simplest amino acid, froze it onto a dust grain like you'd find in space, and zapped it with energy that mimics cosmic radiation. They found that the amino acids linked up to form a two-brick chain, the first step towards building a protein. This means the essential first chains for life could be forming all over space and delivered to new planets by comets and asteroids.