AlphaGenome author roundtable

AI 总结

• 视频由产品经理、基因组负责人和论文一作共同讲述AlphaGenome的研发背景与目标。
• AlphaGenome定位为统一的DNA序列到功能模型，重点解决人类基因组98%非编码区域的功能解读难题。
• 团队强调其在遗传变异功能影响预测上的高准确性，可支持科学家更快筛选关键变异。
• 内容介绍了在TPU上处理长序列高分辨率建模的工程突破，以及对剪接、接触图等复杂生物过程的建模。
• 项目已开放API，便于研究者快速进行变异打分，以推动疾病机制理解与后续研究协作。

内容摘录

Join the Google DeepMind Genomics team for a deep dive into AlphaGenome. In this video, Dhavi (Product Manager) is joined by Ziga (Genomics Lead) and AlphaGenome first authors Natasha, Jun, and Tom to share the story behind AlphaGenome, their unified DNA sequence-to-function model recently published in Nature.

They discuss how the team built a system useful for decoding the 98% of the human genome that is non-coding - helping scientists predict the functional impact of genetic variants with unprecedented accuracy.

Plus, hear about the engineering breakthroughs required to process long sequences at high resolution on TPUs, model complex biological processes like splicing and contact maps, and how the API enables researchers to rapidly score variants for accelerating disease understanding.

Read the paper https://www.nature.com/articles/s41586-025-10014-0 and join the AlphaGenome community https://www.alphagenomecommunity.com/

Chapters
00:00 - Intros
00:39 - Part 1 The Big Picture: Why AlphaGenome?
06:28 - Part 2 The Model: What’s new and how we built it
19:26 - Part 3 From Model to API: Opening it up
23:29 - Part 4 The Future: What’s next and how to get involved