How to design enzymes with AI

TL;DR

AI enables designing new enzymes from scratch for targeted reactions, plus tools to predict DNA mutation effects and generate realistic motions. It showcases world-building, motion generation, interactive AI UIs, and open source models for video, images, and 3D assets, highlighting a period of rapid AI tool expansion. It emphasizes practical access via GitHub and real-time interaction capabilities.

Transcript

AI never sleeps and this week has been absolutely insane. Google DeepMind drops an AI that can design new enzymes from scratch based on your specified purpose. They also released another AI model that can figure out the effects of DNA mutations in plants. OpenAI drops not one, not two, but over 700 breakthroughs on math problems that we haven't bee... Read More

Key Insights

• WorldPlay 2 is an open source world generator that runs locally and preserves memory for long term consistency, enabling interactive character and object prompts and effects
• MSFlow is a motion space flow system that generates human motion from text prompts and lets you constrain joints, with training and evaluation scripts and a compact 1.1 GB size on Hugging Face under the Apache 2 license
• Kandinsky 6.0 provides two models, a lighter 3.8 billion parameter variant and a larger 29 billion parameter pro variant, plus a super resolution latent upscaler to enhance outputs
• OpenAI introduces intelligent UI that builds interactive diagrams and cost plans inside responses, and GPT-6 like behavior to start answering while still thinking, improving response speed
• ALoDLM and Kandinsky 6.0 are available through Hugging Face and diffusers ecosystems, making open source diffusion based tools more accessible for users to test
• AlphaProtein Novo demonstrates AI designed enzymes from scratch and uses AlphaFold 3 for structural prediction, marking a potential shift in enzyme design workflows
• Botanic 1 is a living model by DeepMind for open world interactions, expanding the scope of model-based living environments in the AI space
• Looped diffusion language model by Amazon combines diffusion models with large language models and multiple loops to refine answers, achieving higher throughput and lower compute under a CC non-commercial license

Questions & Answers

Q: A real search query question about the video (e.g. 'How to...', 'What is...', 'Why does...', 'When should...'). Question 1

How can AI design new enzymes from scratch and what tools support this process? The Alpha Protein Novo project demonstrates designing both the 3D structure and amino acid sequence of a novel enzyme, followed by AlphaFold 3 based prediction of folding and interactions. This approach could speed up developing enzymes for new chemical reactions and environmental applications, though the work remains early stage and not yet peer reviewed.

Q: What is WorldPlay 2 and how does it differ from previous world generators?

WorldPlay 2 is an open source world generator that runs locally, allowing real time interaction with characters and objects, and it can add effects or animations on demand. It remembers past actions to maintain consistency over long sessions, and the code is released on GitHub for users to download, set up, and run on their own hardware.

Q: What capabilities does MSFlow offer for motion generation?

MSFlow is a motion space flow system that generates human motion from text prompts and lets users constrain joints to fixed positions at given times. It ships with training and evaluation scripts and a compact total size suitable for consumer GPUs, enabling precise control over generated movements and timing for animations.

Q: How does Kandinsky 6.0 compare between its light and pro models?

Kandinsky 6.0 offers a smaller light model at 3.8 billion parameters and a pro model at 29 billion parameters. The lightweight version supports five second clips and a text to video capability, while the pro version aims for higher quality outputs. A latent super resolution model can upscale outputs by up to four times.

Q: What is intelligent UI in OpenAI tools and what does it enable?

Intelligent UI enables ChatGPT to generate interactive interfaces rather than plain text. For example, it can create interactive diagrams to explain a bike, or plan a Sunday lamb roast with cost breakdowns and checklists. It speeds up responses by starting while reasoning, improving usefulness and speed for complex tasks.

Q: What is the looped diffusion language model and what are its claimed benefits?

The looped diffusion language model from Amazon combines diffusion model ideas with large language models and iterative loops to refine answers. It can predict multiple parts of an answer in parallel, achieving higher throughput and lower compute. The model has smaller and larger variants, released under a Creative Commons non-commercial license.

Q: What is Alpha Protein Novo trying to achieve and why is it significant?

Alpha Protein Novo aims to design entirely new enzymes from scratch for specified chemical reactions, reducing the need to search natural enzymes. By integrating AI design with structure prediction tools, it could accelerate development of enzymes for pollutant breakdown or efficient chemical production, marking a hopeful step toward more programmable biochemistry.

Q: How do open source releases impact AI tool accessibility according to the video?

Open source releases like WorldPlay 2, MSFlow, Kandinsky 6.0 components, and intelligent UI make advanced AI tools more accessible, enabling researchers and hobbyists to download, run locally, modify prompts, and experiment with open models. Access via GitHub and Hugging Face lowers barriers to experimentation and cross-platform integration across video, image, and motion domains.

Summary & Key Takeaways

• WorldPlay 2 is an open source world generator that runs locally, maintains memory and consistency across sessions, and supports interactive prompts for animations and effects.

• Kandinsky 6.0 offers light and pro models for text to video with a new super resolution option, accessible via diffusers and comfy UI, though clip length may be limited in the lightweight version.

• Intelligent UI in ChatGPT enables interactive diagrams, cost breakdowns, and checklists within responses, expanding beyond plain text to more visual and actionable outputs.