The End Of Knowledge - Vault 5/95 - CVPR - 2024 - Design of New Protein Functions Using Deep Learning

graph LR classDef main fill:#f9d4d4, font-weight:bold, font-size:14px classDef design fill:#d4f9d4, font-weight:bold, font-size:14px classDef models fill:#d4d4f9, font-weight:bold, font-size:14px classDef applications fill:#f9f9d4, font-weight:bold, font-size:14px classDef properties fill:#f9d4f9, font-weight:bold, font-size:14px classDef techniques fill:#d4f9f9, font-weight:bold, font-size:14px Main[Design of New
Protein Functions Using
Deep Learning] Main --> Design[Protein Design] Main --> Models[AI Models] Main --> Applications[Applications] Main --> Properties[Protein Properties] Main --> Techniques[Design Techniques] Design --> A[Creating novel proteins
computationally using deep learning 1] Design --> B[Designing proteins with
multiple conformational states 22] Design --> C[New catalysts for
various applications 17] Design --> D[Proteins template growth
of inorganic materials 12] Design --> E[Proteins functionalize carbon nanotubes 13] Design --> F[Designing RNA with
specific structures, functions 16] Models --> G[Deep learning model
predicts protein structures 2] Models --> H[Diffusion model generates
protein structures, sequences 3] Models --> I[Neural network designs
amino acid sequences 4] Models --> J[Diffusion models create
small molecule binders 15] Models --> K[Exploring latent space
for protein diffusion 29] Applications --> L[Designed proteins target
various medical conditions 5] Applications --> M[Proteins neutralize snake
venom toxins 6] Applications --> N[Synthetic antibodies mimic
immune system components 7] Applications --> O[Cylindrical proteins function
as molecular channels 8] Applications --> P[Proteins bind specific
DNA sequences 9] Applications --> Q[Self-assembling proteins for
drug delivery 10] Properties --> R[De novo proteins
exhibit high stability 20] Properties --> S[Improved stability reduces
cold storage need 23] Properties --> T[Flexible building blocks
create larger assemblies 11] Techniques --> U[Technique specifies catalytic
site geometry 18] Techniques --> V[Proteins mimic photosystems
for light harvesting 19] Techniques --> W[Applying computer vision
to protein design 24] Techniques --> X[Considering sequences and
3D coordinates simultaneously 27] Techniques --> Y[Applying diffusion models
to 3D structures 28] Techniques --> Z[Applying reinforcement learning
to protein design 30] Applications --> AA[Protein networks compute
within cells 14] Applications --> AB[Protein-based vaccines protect
multiple virus strains 21] Techniques --> AC[Computationally generated data
augments structure databases 25] Applications --> AD[Interacting proteins perform
cellular computations 26] class Main main class Design,A,B,C,D,E,F design class Models,G,H,I,J,K models class Applications,L,M,N,O,P,Q,AA,AB,AD applications class Properties,R,S,T properties class Techniques,U,V,W,X,Y,Z,AC techniques

Resume:

1.- De novo protein design: Creating brand new proteins from scratch using computational methods, particularly deep learning techniques.

2.- RosettaFold: A deep learning model developed to predict protein structures from amino acid sequences.

3.- RF diffusion: A diffusion model for generating new protein structures and sequences.

4.- Protein MPNN: A message-passing neural network for designing amino acid sequences that fold into specific structures.

5.- Medical applications: Using designed proteins to target diseases like autoimmune disorders, cancer, and neurodegenerative conditions.

6.- Snake venom antidote: Designing proteins that bind to and neutralize snake venom toxins.

7.- Antibody design: Creating synthetic antibodies for therapeutic use, mimicking natural immune system components.

8.- Protein channels: Designing cylindrical proteins that can function as ion channels or molecular sensors.

9.- DNA-binding proteins: Creating proteins that recognize and bind to specific DNA sequences for potential gene editing applications.

10.- Nanoparticle design: Developing self-assembling protein structures for vaccine platforms and drug delivery.

11.- Quasi-symmetry: A method to create larger protein assemblies by introducing flexibility in building blocks.

12.- Mineralization: Designing proteins that can template the growth of inorganic materials like zinc oxide.

13.- Carbon nanotube binding: Creating proteins that can wrap around and functionalize carbon nanotubes.

14.- Protein-based computing: Designing protein networks that can perform simple computations within cells.

15.- Small molecule binding: Using diffusion models to create proteins that bind specific small molecules.

16.- RNA design: Applying diffusion models to design RNA molecules with specific structures and functions.

17.- Enzyme design: Creating new catalysts for various applications, including plastic degradation and green chemistry.

18.- Guideposting: A technique to specify catalytic site geometry during the protein design process.

19.- Artificial photosynthesis: Designing proteins that mimic natural photosystems for light harvesting and electron transfer.

20.- Protein stability: De novo designed proteins often exhibit high thermal stability compared to natural proteins.

21.- Universal vaccines: Developing protein-based vaccines that could protect against multiple strains or families of viruses.

22.- Protein dynamics: Designing proteins with multiple conformational states or continuous motions for specific functions.

23.- Shelf stability: De novo designed proteins often have improved stability, potentially reducing the need for cold chain storage.

24.- Computer vision in biology: Applying techniques from computer vision and image processing to protein design and analysis.

25.- Synthetic training data: Exploring the use of computationally generated data to augment experimental protein structure databases.

26.- Protein programming: Creating networks of interacting proteins to perform computations or make decisions within cells.

27.- Multimodal protein representation: Considering both amino acid sequences and 3D coordinates in protein diffusion models.

28.- Structure diffusion: Applying diffusion models directly to 3D protein structures rather than sequences.

29.- Latent diffusion: Exploring latent space representations for simultaneous diffusion of protein sequence and structure.

30.- Reinforcement learning: Applying techniques like DPO (Direct Preference Optimization) to protein design tasks.

Knowledge Vault built byDavid Vivancos 2024