Folding Forum

Hello to all Folding@home users!

I would like to share some thoughts with you about the potential of artificial intelligence (AI) in integrating with the Folding@home project. As you know, Folding@home has proven to be an extraordinary initiative in contributing to the understanding of diseases and the discovery of new therapeutic approaches. However, integrating AI could bring additional benefits and accelerate scientific research progress.

AI is a powerful tool in data analysis and simulation of biological processes. Here's how it could be easily integrated into the Folding@home project:

Data analysis: AI could be used to analyze the vast amount of data generated by Folding@home. With their ability to recognize complex patterns and correlations, AI could help identify relevant protein structures for disease understanding and discover new therapeutic targets.

Protein structure prediction: AI could be employed to improve predictions of protein folding. Through machine learning and advanced algorithms, AI could contribute to obtaining more accurate results and speeding up the process of identifying the three-dimensional structure of proteins.

Virtual screening of molecules: AI could perform virtual screenings of large libraries of molecules to identify compounds that may have therapeutic activity. This approach could expedite the discovery of new drugs or more effective therapeutic combinations.

Optimization of computational processes: AI could be utilized to optimize the allocation of Folding@home's computational resources. This could enable more efficient utilization of available computing capabilities and reduce the time required to complete simulations.

Integrating AI into Folding@home would require collaboration between experts in artificial intelligence, biology, and medicine. However, I believe that this synergy can lead to significant scientific discoveries and accelerate research towards understanding and treating diseases.

I am excited to hear your thoughts and opinions on how AI could be integrated into Folding@home. What do you think about this possibility? Do you have any ideas on how it could be realized? I look forward to reading your comments and continuing to support this important scientific initiative together.

Thank you all for your commitment to Folding@home!

Manuele Friargiu (Italy)

(This is just my understanding, I am no expert)
Most of the F@H math is 32 bit Floating Point, with occasional 64 bit FP doing accuracy checks. The new tensor cores used in AI are 8 bit FP. This is not enough accuracy for the repetitive calculations done in F@H. It may help in the logic part of the program, but it not the right sort of math for the repetitive calculations. I hold out more hope for AVX512 math.

Hola,

Like Jimbo I'm not claiming to be an expert, so just my thoughts here...

When you built a house, you don't integrate the concrete mixer with the hammer and saw, you need them all independently to build a better house. Maybe AI and F@h are like that. As far as I understand, F@h is primarily used to understand how technically proteins are made and interact - the actual mechanics of the proteins. Like: how does the COVID virus actually get hold of the cells it infects. I believe AI is used more to look for proteins that have the most potential (caveat: AI is trained on what we already know, not on what we don't know; there is still so much we don't know we may miss out on quite a bit if we blindly use AI). AI looks for patterns, deviations, etc, and AI can generate based on similarities with a training set. AI however does not understand the how or why.

So AI could be useful to focus efforts in specific projects, and possibly take the results from F@h research for a next step. On the other hand, F@h is also used to do fundamental research, in the realm of what we don't know yet.

While it's true that tensor cores in AI are primarily designed for lower precision calculations like 8-bit FP, they can still be valuable in certain aspects of Folding@home. AI can assist in optimizing algorithms, improving resource utilization, and addressing complex problems related to biological processes.
The integration of AI and Folding@home can be seen as a synergistic approach. AI can aid in the logical and optimization aspects of the program, while the existing computational resources handle the repetitive calculations. By combining their strengths, a more comprehensive and efficient solution can be achieved.
Thank you for reading and your interesting answer

Manuele (Italia)

JimboPalmer wrote: ↑Wed Aug 09, 2023 3:18 am (This is just my understanding, I am no expert)
Most of the F@H math is 32 bit Floating Point, with occasional 64 bit FP doing accuracy checks. The new tensor cores used in AI are 8 bit FP. This is not enough accuracy for the repetitive calculations done in F@H. It may help in the logic part of the program, but it not the right sort of math for the repetitive calculations. I hold out more hope for AVX512 math.

You're right, thank you for your point of view not negligible. AI can indeed be used to search for proteins with high potential based on patterns and similarities found in the training data. The integration of AI and Folding@home can be seen as a collaborative approach where each tool plays a unique role. Folding@home's fundamental research can provide the necessary foundation for understanding proteins, while AI can assist in identifying patterns and focusing efforts on specific projects. By leveraging the strengths of both approaches, we can potentially enhance our understanding of proteins and contribute to advancements in scientific research.

Manuele (Italia)

PaulTV wrote: ↑Wed Aug 09, 2023 9:27 am Hola,

Like Jimbo I'm not claiming to be an expert, so just my thoughts here...

When you built a house, you don't integrate the concrete mixer with the hammer and saw, you need them all independently to build a better house. Maybe AI and F@h are like that. As far as I understand, F@h is primarily used to understand how technically proteins are made and interact - the actual mechanics of the proteins. Like: how does the COVID virus actually get hold of the cells it infects. I believe AI is used more to look for proteins that have the most potential (caveat: AI is trained on what we already know, not on what we don't know; there is still so much we don't know we may miss out on quite a bit if we blindly use AI). AI looks for patterns, deviations, etc, and AI can generate based on similarities with a training set. AI however does not understand the how or why.

So AI could be useful to focus efforts in specific projects, and possibly take the results from F@h research for a next step. On the other hand, F@h is also used to do fundamental research, in the realm of what we don't know yet.