Predicting protein function with ProtNLM @GoogleResearch

Uploaded March 2023 | Updated September 2026, 2 weeks ago

In this episode of ResearchBytes, you’ll learn how Google Research is working together with the European Bioinformatics Institute at the European Molecular Biology Laboratory to predict the function of proteins. Andreea Gane from the Brain Team at Google Research shares insights on ProtNLM, which can predict a short functional description from a protein’s amino acid sequence. Predictions from ProtNLM are now used to label millions of protein sequences that previously had no description in the institute’s database.

Resources:
Read the ProtNLM preprint → storage.googleapis.com/brain-genomics-public/research/proteins/protnlm/uniprot_2022_04/protnlm_preprint_draft.pdf

Read the ProtNLM help page on the UniProt website → uniprot.org/help/ProtNLM

See ProtNLM predictions on the UniProt website → uniprot.org/uniprotkb?query=(source:google)

Explore the Colab notebook that provides evidence for ProtNLM predictions → colab.research.google.com/github/google-research/google-research/blob/master/protnlm/protnlm_evidencer_uniprot_2023_01.ipynb

Explore the curation example shown in the video, starting with the UniProt entry containing the ProtNLM prediction. → uniprot.org/uniprotkb/D2GX75/entry#names_and_taxonomy

Chapters:

0:00 - Intro
0:50 - What’s the challenge?
3:16 - What did we do?
3:40 - How did we do it?
5:00 - What did we observe?
6:10 - What’s next?

We thank our collaborators at EMBL-EBI, and in particular Dr. Elena Speretta for assembling the curation example included in the video, and for the careful work in evaluating predictions.
ebi.ac.uk/people/person/elena-speretta

Watch more episodes of ResearchBytes → https://goo.gle/ResearchBytes
Subscribe to the Google Research Channel → https://goo.gle/GoogleResearch

#ResearchBytes

Predicting protein function with ProtNLM

Fighting extreme heat in urban environments

Research@ Bangalore: A conversation with Jeff Dean

From satellite imagery to queryable insights l Google Earth AI

How we built two new brain maps using Google AI

Building an ML model to accurately estimate the reflectivity of roofs

Research@ NYC: A Fireside Chat with Corinna Cortes & Katherine Chou

Scaling Language-Image Learning in 100 Languages with PaLI

$Making Weather Forecasts More Accurate Climate change is real, and its effects are becoming increasingly evident. Traditional climate models struggle to produce clear, accurate pictures of how our climate will continue to change due to limitations in how the models represent complex phenomena. This video explores NeuralGCM, a groundbreaking AI-powered approach developed by Google Research that could someday offer a faster, more efficient, and more accurate way to predict climate change. Learn more about NeuralGCM in Nature! You’ll hear from: Stephan Hoyer, PhD, AI weather/climate lead at Google Research Janni Yuval, PhD, visiting Google Research Scientist Watch more Research Bytes → https://goo.gle/ResearchBytes Subscribe to the Google Research Channel → https://goo.gle/GoogleResearch The Challenge of Climate Prediction Accurate climate predictions are a matter of urgency because they could allow us to prepare and adapt to climate change. NeuralGCM tackles the limitations of current physics-based models by combining them with machine learning. This allows for faster simulations and more accurate results. How Does NeuralGCM Work? Traditional climate models divide the world into large cubes, simulating the climate in each and sometimes making approximations about small-scale phenomena like clouds. NeuralGCM uses AI to make better approximations by learning about these phenomena from past data patterns, ultimately leading to a more accurate prediction engine. A Game-Changer for Climate Science We’ll dive into the revolutionary potential of NeuralGCM. It is more accurate than traditional models, reduces bias, and achieves supercomputer-level accuracy while using a fraction of the computational power. This allows researchers to simulate thousands of days in the same amount of time it would take them to simulate just one day using a high-resolution physics model, significantly improving our understanding of future climate scenarios. Benefits for Humanity NeuralGCM is a groundbreaking step toward a new approach of building climate models that could make them faster and more accurate. Improving climate models would empower us to understand and prepare for the impacts of climate change. We can mitigate global warming by reducing carbon emissions and adapt to its effects through regional planning and infrastructure development. Open-source research like NeuralGCM supports Googles commitment to making this technology universally accessible. This video is perfect for anyone interested in: Climate change and its impact Artificial intelligence applications The future of weather prediction Technological solutions for a sustainable future #AI #climateaction #GoogleResearch #climatescience #machinelearning #AIforsocialgood #globalwarming #climatestrategy #ECMWF #naturemagazine$

How to get started with the Google FloodHub

Project Relate - Communication made easier