Anti-Microbial Proteins Are Being Developed With AI By... Salesforce? (neowin.net) 13
segaboy81 shares a report from Neowin:
What do you get when the world's largest CRM breaks into the research industry and leverages AI to build their products? You get ProGen, a new AI system that can make artificial enzymes from scratch that can work just as well as real ones found in nature. ProGen was made by Salesforce Research (yes, that Salesforce) and uses language processing to learn about biology. In short, ProGen takes amino acid sequences and turns them into proteins....
"The artificial designs are better than ones made by the normal process," said James Fraser, a scientist involved in the project. "We can now make specific types of enzymes, like ones that work well in hot temperatures or acid."
To make ProGen, the scientists at Salesforce fed the system amino acid sequences from 280 million different proteins. The AI system quickly made a staggering one million protein sequences, of which 100 were picked to test. Out of these, five were made into actual proteins and tested in cells. That's just 0.0005% of the generated results....
The code for ProGen is available on Github for anyone who wants to try it (or add to it)
The project shows "how generative AI can lead to potential solutions for addressing challenges in human disease and the environment," argues a statement form Salesforce.
More details from New Scientist: The AI, called ProGen, works in a similar way to AIs that can generate text. ProGen learned how to generate new proteins by learning the grammar of how amino acids combine to form 280 million existing proteins. Instead of the researchers choosing a topic for the AI to write about, they could specify a group of similar proteins for it to focus on. In this case, they chose a group of proteins with antimicrobial activity.
The researchers programmed checks into the AI's process so it wouldn't produce amino acid "gibberish", but they also tested a sample of the AI-proposed molecules in real cells. Of the 100 molecules they physically created, 66 participated in chemical reactions similar to those of natural proteins that destroy bacteria in egg whites and saliva. This suggested that these new proteins could also kill bacteria. The researchers selected the five proteins with the most intense reactions and added them to a sample of Escherichia coli bacteria. Two of the proteins destroyed the bacteria.
The researchers then imaged them with X-rays. Even though their amino acid sequences were up to 30% different from any existing proteins, their shapes almost matched naturally occurring proteins. James Fraser at the University of California, San Francisco, who was part of the team, says it was not clear from the outset that the AI could work out how to change the amino acid sequence so much and still produce the correct shape.... He was surprised to have found a well-functioning protein in the first relatively small fraction of all the ProGen-generated proteins that they tested.
"The artificial designs are better than ones made by the normal process," said James Fraser, a scientist involved in the project. "We can now make specific types of enzymes, like ones that work well in hot temperatures or acid."
To make ProGen, the scientists at Salesforce fed the system amino acid sequences from 280 million different proteins. The AI system quickly made a staggering one million protein sequences, of which 100 were picked to test. Out of these, five were made into actual proteins and tested in cells. That's just 0.0005% of the generated results....
The code for ProGen is available on Github for anyone who wants to try it (or add to it)
The project shows "how generative AI can lead to potential solutions for addressing challenges in human disease and the environment," argues a statement form Salesforce.
More details from New Scientist: The AI, called ProGen, works in a similar way to AIs that can generate text. ProGen learned how to generate new proteins by learning the grammar of how amino acids combine to form 280 million existing proteins. Instead of the researchers choosing a topic for the AI to write about, they could specify a group of similar proteins for it to focus on. In this case, they chose a group of proteins with antimicrobial activity.
The researchers programmed checks into the AI's process so it wouldn't produce amino acid "gibberish", but they also tested a sample of the AI-proposed molecules in real cells. Of the 100 molecules they physically created, 66 participated in chemical reactions similar to those of natural proteins that destroy bacteria in egg whites and saliva. This suggested that these new proteins could also kill bacteria. The researchers selected the five proteins with the most intense reactions and added them to a sample of Escherichia coli bacteria. Two of the proteins destroyed the bacteria.
The researchers then imaged them with X-rays. Even though their amino acid sequences were up to 30% different from any existing proteins, their shapes almost matched naturally occurring proteins. James Fraser at the University of California, San Francisco, who was part of the team, says it was not clear from the outset that the AI could work out how to change the amino acid sequence so much and still produce the correct shape.... He was surprised to have found a well-functioning protein in the first relatively small fraction of all the ProGen-generated proteins that they tested.
Question about this AI (Score:3)
This specific item was designed/used to create anti-microbial proteins by referencing those millions of amino acids. Would there be limits on what this AI could research? By that I mean, would there be exclusions because we told it this combination or that would be impossible so it shouldn't consider it?
clean (Score:2)
There is only clean advertising through Salesforce. This spam guaranteed to be microbe free.
An excellent pairing (Score:2)
I'f you've ever worked with Salesforce APIs, you can understand why they would be researching the strongest possible disinfectant to submerge yourself in afterward.
Given we're talking about SalesForce... (Score:3)
Should we actually believe anything they claim?
Tick (Score:2)
Put the clock forward
As a biologists (Score:2, Interesting)
I worked in protein design for a few years as a project scientists. Thinking that just reading literature text will help you is grossly naive. Most literature is wrong (but we dont know which ones) and the rest is only comprehensible when properly interpreting the manuscripts figures.
I looked at all these ai generated protein folds, and for the simple cases these were correct. However most proteins come in complexes and there the prediction were not above random noise
So good luck salesforce, you will come u
Re: (Score:2)
I can imagine use cases where this sort of 'bunch of leads and a ranking on where to burn money' is helpful. Just like I can see times where a AI-generated document or code is easier to fix than for me to write from scratch.
CNC is often 'clone this machinist's actions' and iterative adjustments/improvements, then it just runs that 1000x or whatever. I'd rather have an AI-start on process documentation that I quickly edit/fix than to continue lying to myself 'I"ll write it soon." The code and documents I'
Re: (Score:2)
Fully agree, thanks
'Now I am become Death, the destroyer of worlds' (Score:2)
A nuke requires building them, and then actual delivery. Plenty of time for morals to kick in (disregarding Russia's doom's day device with that one).
But an AI, that is connected to the net, could potentially go far in thinking and without a shred of morals (much like today's CEOs around the world).
This is something to pay attention to.
Re: (Score:1)
understandable (Score:2)