AI Launches a Fierce Assault on Cancer, Virtually Tests 4,000 Anti-Cancer Drugs, Google Gemma Downloads Exceed 1 Billion
Google's open-source Gemma AI model family has reached a new milestone, surpassing 10 billion cumulative downloads in two and a half years since its February 2024 launch. This growth has fueled a vast ecosystem, with over 100,000 developer-derived model variants powering diverse applications, from NASA satellites to over 100 million phones.
A standout application involves cancer research. In collaboration with Google Research and Yale, a team developed a 27-billion-parameter model called Cell2Sentence-Scale 27B (C2S-Scale). This model innovatively converts complex single-cell sequencing data into a "sentence" of gene names ranked by activity, allowing it to analyze cellular behavior. It was used to conduct a "dual-context virtual screen" of over 4,000 drugs, aiming to find one that specifically boosts antigen presentation—a key process for immune system recognition—only in the presence of low interferon signals (a "cold tumor" context), not in neutral conditions.
The model identified silmitasertib (CX-4945), a CK2 kinase inhibitor, as a top candidate. Subsequent live-cell experiments validated this AI-generated hypothesis, showing that the drug combined with low-dose interferon increased antigen presentation by about 50%, whereas neither element alone had a significant effect. This represents a novel, AI-proposed mechanistic pathway for potential cancer therapy.
Beyond biomedicine, the Gemma ecosystem is expanding into unique domains. Models have been deployed in space on high-performance GPUs and are being used in projects like DolphinGemma to decipher dolphin communication patterns.
The 10-billion-download milestone underscores a key trend: the disruptive potential of AI lies not just in raw model capability but in its widespread, silent integration into diverse hardware and real-world applications, as evidenced by the thriving "Gemmaverse" community.
marsbit08/21 08:06
