AlphaGenome Atlas: Understanding the human genome

Summary
This video presents AlphaGenome Atlas, a platform and dataset developed by Google DeepMind to predict the functional consequences of genetic mutations across the human genome. Google DeepMind researchers Pushmeet Kohli, Žiga Avsec, Jun Cheng, and Dhavi Hariharan present the project alongside academic genomics researchers Dr. Gareth Hawkes (University of Exeter) and Sam Bryen (Centre for Population Genomics).

What is shown

• Pushmeet Kohli discussing decoding the genome as understanding "the language of life" [00:00].
• Conceptual animations of DNA sequences, base pairs, and molecular variation networks [00:11, 00:28, 00:51, 01:08].
• Dr. Gareth Hawkes and Sam Bryen explaining the scale of whole-genome sequencing and variant interpretation [00:17, 00:43].
• Žiga Avsec explaining how AlphaGenome evaluates genomic sequences and condenses thousands of predictions into a single AVI (variant impact) score [01:05, 01:31, 01:50].
• The AlphaGenome Atlas web portal interface, featuring searches for specific genetic variants (e.g., chr22:36201698:A>C), Phred scores, feature importance breakdowns (splicing, CAGE, ChIP-Histone, Cactus), and genomic locus track visualizations [02:20 - 02:30].
• An interactive multi-agent life-sciences interface displayed on a laptop, integrating AlphaFold and AlphaGenome for pathogenicity analysis [03:22].

Claims & numbers

• The human genome contains approximately 3 billion base pairs of information, which would take tens of years to read at one second per base pair (Dr. Gareth Hawkes, Sam Bryen) [00:27, 00:39, 00:53].
• AlphaGenome condenses roughly 10,000 predicted features per variant into a single unified AVI (variant impact) score (Žiga Avsec) [01:31, 01:54].
• Google DeepMind precomputed AVI scores for all 9 billion possible single-nucleotide changes in the human genome (Žiga Avsec) [01:45].
• AlphaGenome Atlas encompasses approximately 1 petabyte of genomic prediction data made available through a web portal for biologists (Dhavi Hariharan) [02:18].
• Google DeepMind is building a broader agentic framework linking models such as AlphaFold and AlphaGenome to assist scientists in solving complex research problems (Pushmeet Kohli) [03:15].

Notable quotes

• "AlphaGenome is a model that looks at chunk of the genome sequence and predicts what happens if you make a single mutation to this region." — Žiga Avsec [01:05]
• "We have precomputed this for all 9 billion possible single-letter changes in the human genome." — Žiga Avsec [01:45]
• "AlphaGenome Atlas is probably the most comprehensive and accurate variant effect prediction platform. It's about 1 petabyte of data." — Dhavi Hariharan [02:12]

Assessment
This is an official launch overview video combining researcher interviews, conceptual 3D biological animations, and real screencasts of the AlphaGenome Atlas genome browser UI. The video clearly illustrates the platform's core interface and precomputed variant scoring system, concluding with a brief look at an experimental agentic scientific research environment.

Described by gemini-3.8-flash on 2026-10-06 from the video's audio and frames.