Looking at the updates in the latest version, you’ll notice that Alphafold 3 expands its modeling capabilities to a wider range of molecular structures, such as ligands (ions or molecules with specific binding properties), other ions, and what’s referred to as “post-translational modifications” – here’s the Wikipedia explanation.

In addition, Alphafold 3 utilizes an updated “Pairformer” architecture for handling pairwise relationships (more on this later) – delivering enhanced prediction accuracy and better performance in certain types of predictions. You can find additional details from NIH.

The original Alphafold system earned its creators, John Jumper and Demis Hassabis, a Nobel prize, and these tools continue to reshape the landscape of drug discovery.

Practical Applications

So how does Alphafold assist major pharmaceutical companies?

In a TED talk discussing some of its commercial success, Lauren Davis, who has connections to MIT, gives us a glimpse into how the technology is used to help develop life-saving drugs.

Davis highlights a “transformative” process where new tools allow for faster development within a more efficient framework. One key part of this, she explains, is target identification – predicting the structure of a particular target. This enables companies to bypass some of the costly and labor-intensive human and animal testing phases, which are also ethically complex.

She likens the system to a dating app, where potential matches are filtered before actual testing takes place – similar to going on a date after narrowing down options (i.e., investing time and energy).

She also discusses the scoring of possible inhibitors, a feature she finds exciting – especially since she was once part of the MIT soccer team.

Overall, Davis illustrates how Alphafold functions in the real, commercial world.

Is Alphafold 3 Open Source?

That covers some of how Alphafold contributes to the medical field.

But there’s another aspect of this new model I wanted to explore, and I received conflicting responses from different LLM models, so let’s dive into that:

If you ask ChatGPT whether Alphafold 3 is open source, here’s what you get – at least in my experience:

“(Alphafold 3 is) not open-sourced; instead, it is available through a cloud-based platform provided by DeepMind for non-commercial research purposes. This approach has raised concerns in the scientific community regarding transparency and accessibility.”

As a reference, the model points to this Wired article.

However, when I reviewed a post from Dario Amodei from last November, I found a brief but clear statement suggesting that Alphafold 3 has become open source:

“AI protein prediction tool, AlphaFold3, is open source.”

When I asked Copilot the same question via a Bing search, the response was:

“Not fully open source - AlphaFold 3 is not fully open source. While the source code and model weights are accessible for academic use under certain non-commercial restrictions, access to the model weights is restricted to those affiliated with academic institutions. This means researchers can use the software for non-commercial purposes, but they cannot freely access the training weights for commercial applications.”

Several sources were cited in support of this view.

So, based on all this information, the most probable conclusion is that Alphafold 3 is “partially” open source – as Copilot put it, some components like the weights are publicly available, while others remain proprietary.

ChatGPT, meanwhile, argues that the cloud-based platform hosting Alphafold 3 is “maintained by DeepMind for non-commercial research purposes.”

Regardless, we don’t have to speculate here: simply check the GitHub page for the latest version.

This situation, however, demonstrates how different AI models can provide varying responses to the same query – something we must learn to navigate.

Doing the Work

In any case, this updated version of Alphafold continues the legacy of providing powerful tools for drug discovery. In her talk, Davis cites lisinopril as an ACE inhibitor, explaining how Alphafold can be applied to predict and model how an inhibitor behaves.

In practice, millions of Americans take medications like these, which are used across a wide range of health conditions, making AI-assisted development highly valuable.

Will this reduce drug costs? That remains to be seen.

The above is the detailed content of Alphafold 3 Extends Modeling Capacity To More Biological Targets. For more information, please follow other related articles on the PHP Chinese website!