In just two days, the world of AI seems to have hit the “accelerator pedal.”
First, yesterday, Google CEO Sundar Pichai announced on X (formerly Twitter) that they had partnered with Yale University to develop a new model based on the Gemma architecture. This model makes it easier for the immune system to recognize cancer cells.
Following up immediately today, Google Research delivered another heavy blow: DeepSomatic, which can identify genetic mutations in tumors with unprecedented precision.
Someone on X commented:
“Two AI breakthroughs against cancer in two days. 2025 will probably be humanity's last 'ordinary year.'
The world thereafter will accelerate rapidly.”
It might sound exaggerated, but I truly feel that the wheels of history are turning.
Gemma: The First Time AI "Understood Cell Language"
The new model code-named C2S-Scale 27B, released by Google in collaboration with Yale University, is not a language model that “writes text,” but a model that “reads cells.”
Simply put, it can read the state of every cell like reading a sentence.
The research team tasked it with analyzing thousands of drugs to see which ones could make cancer cells more easily recognizable by the immune system. The result—AI actually discovered a new lead. An old drug called Silmitasertib (a CK2 kinase inhibitor) was predicted by the model to have the potential to “heat up the tumor.”
What does that mean?
There is a type of “cold tumor” in cancer that the immune system cannot recognize at all. But this drug, combined with faint immune signals, can make tumor cells light up a “signal light”—telling the immune system: “I'm here!”
Experimental verification was also successful.
When scientists treated cells in the lab with Silmitasertib + interferon, the cell's antigen presentation capability increased by 50%, meaning it was easier for the immune system to detect and attack.
DeepSomatic: The Microscope That Finds "Typos" in Genes
Today, Google delivered the second blockbuster: DeepSomatic. This is an AI that can detect cancerous gene mutations with unprecedented accuracy, marking a huge leap toward humanity's goal of “curing cancer.”
Cancer is essentially the “instruction code going wrong” in the body: the genetic program for cell division is hijacked, causing cells to multiply wildly, evade immunity, and constantly mutate and spread.
DeepSomatic's mission is to find these erroneous “code fragments.”
DeepSomatic uses machine learning and convolutional neural networks to identify these dangerous mutations in tumor DNA—even in complex cases where traditional methods often fail.
It can operate across all mainstream DNA sequencing platforms and can even identify mutations in cancer types that the model was not specifically trained on.
In testing, DeepSomatic outperformed existing tools significantly, not only finding more true mutations but also drastically reducing false positives.
This research was also formally published in Nature.
Final Thoughts
AI is no longer just writing articles or drawing pictures. It is now participating in humanity's understanding and reshaping of life.
Gemma proposed a new approach to fighting cancer, and DeepSomatic identified cancer's “genetic loopholes.” One is asking “why do we get sick,” and the other is answering “how can we cure it.”
Scientists predict: “By 2030, AI might truly be able to decipher cancer.” This used to sound like a movie line.
But this time, even Google said it seriously.
Perhaps, the first step toward curing cancer has truly been calculated on a GPU.
References
Produced by "AI Fan'er"