Chemistry World Podcast · Chemistry World

AlphaGenome & the RNA world hypothesis | The chemical breakdown podcast

·27 min·2 clips

Google DeepMind's AlphaGenome can predict DNA variations across a million base pairs—how will this reshape genetics research?

This episode of the Chemistry World Podcast, hosted by Mariana Kneppers, explores Google DeepMind's AlphaGenome and the RNA world hypothesis for life's origins. Correspondent Mason Wakely and Features Editor Neil Withers join to analyze these topics. AlphaGenome is a deep learning model that predicts the effects of single base pair variations across DNA sequences up to one million base pairs long. The tool is trained on publicly available human and mouse genome data, allowing it to analyze the vast non-coding regions once called "junk DNA." A key limitation is its difficulty predicting interactions between variations more than 100,000 base pairs apart. Since its preview release in 2025, the model has seen significant uptake, with about 3,000 users from nearly 160 countries making roughly a million requests daily. Potential applications include understanding rare diseases like cystic fibrosis and designing therapeutic antisense oligonucleotides. The discussion compares AlphaGenome to its predecessor, AlphaFold, which predicts protein structures and won a Nobel Prize in 2024. The hosts note that while AlphaFold revolutionized its field, AlphaGenome's impact on genomics might be more niche. The RNA world hypothesis, proposed by pioneers like Francis Crick and Leslie Orgel, suggests RNA preceded DNA as life's primary genetic material. This is partly because RNA is single-stranded and simpler, and the discovery that the ribosome's catalytic site is RNA showed it could self-replicate. A leading chemical theory posits hydrogen cyanide as a foundational molecule, using UV light and a reducing atmosphere to build amino acids and nucleotides through reductive homologation. A surprising insight is that over 98% of the human genome is non-coding but crucially influences gene activity, contradicting outdated "junk DNA" notions. The hosts speculate that recreating early life in a lab remains distant due to the need to combine encapsulation, information storage, and metabolism. They debate whether AI could accelerate origins-of-life research, noting the lack of large datasets comparable to those used for AlphaFold. The episode concludes with a historical segment on the accidental 1943 invention of Silly Putty from boric acid and silicone oil. The tone is conversational and educational, with the hosts breaking down complex science for a general audience. This episode would appeal to listeners interested in AI's applications in biology, genetics, and theoretical origins-of-life chemistry. Those seeking highly technical details or definitive answers to historical mysteries might find it too introductory.

As heard by us

A clear listen on DNA prediction, drug discovery, and the chemistry of life's first steps.

The episode is at its best when it turns to the new deep learning model from Google DeepMind, which can predict the effect of small DNA changes across stretches as long as 1 million base pairs and suggests possible uses in understanding the human genome and designing therapeutic…

Read the full review in PlayNext →

Why you'd press play

If you want a practical read on a new DeepMind model for DNA and a side trip into life's origin story, start here.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Chemistry World Podcast