- Видео 10
- Просмотров 8 857
Kilian Mandon
Германия
Добавлен 30 июн 2024
Unlock the secrets of protein folding and revolutionize how we approach drug design, disease understanding, and biological research. AlphaFold is transforming the field, and you can be a part of this revolution.
We break down the complexities of AlphaFold into manageable steps. From the basics of tensors and machine learning to advanced geometry, we'll guide you through hands-on tutorials and interactive explanations.
Become a pioneer in AI-driven structural biology. Gain the skills to not only use existing tools but to innovate and push the boundaries of what's possible.
We break down the complexities of AlphaFold into manageable steps. From the basics of tensors and machine learning to advanced geometry, we'll guide you through hands-on tutorials and interactive explanations.
Become a pioneer in AI-driven structural biology. Gain the skills to not only use existing tools but to innovate and push the boundaries of what's possible.
AlphaFold Decoded: Full Model (Lesson 9)
For the ninth and final video, we bring together all the components to complete the AlphaFold model. You’ll see how the Input Embedder, Evoformer, Structure Module, and other parts interconnect to predict accurate protein structures. This video covers how the modules interact, how recycling improves predictions, and how AlphaFold iteratively enhances its outputs.
The final Jupyter Notebook ties everything together, allowing you to run the full AlphaFold pipeline and see your implementation in action.
Assignment:
1) Clone github.com/kilianmandon/alphafold-decoded
2) Work on tutorials/model
Setup Instructions on github.com/kilianmandon/alphafold-decoded
Timestamps:
0:00 Intro
0:30 Overview
3:32 In...
The final Jupyter Notebook ties everything together, allowing you to run the full AlphaFold pipeline and see your implementation in action.
Assignment:
1) Clone github.com/kilianmandon/alphafold-decoded
2) Work on tutorials/model
Setup Instructions on github.com/kilianmandon/alphafold-decoded
Timestamps:
0:00 Intro
0:30 Overview
3:32 In...
Просмотров: 318
Видео
AlphaFold Decoded: Structure Model (Lesson 8)
Просмотров 2683 месяца назад
For the eighth video, we focus on the Structure Module, where AlphaFold’s predictions truly come to life. We’ll examine how the Invariant Point Attention (IPA) mechanism refines protein structures by taking Evoformer outputs and iteratively improving backbone translations and rotations. You’ll learn how quaternions and torsion angles are used to adjust atom positions, and we’ll discuss how Alph...
AlphaFold Decoded: 3D Geometry (Lesson 7)
Просмотров 5063 месяца назад
In the seventh video of this series, we cover the core geometric principles that form the basis of AlphaFold’s structure prediction. We’ll explore how AlphaFold translates predicted tensors into atomic positions, using quaternions for backbone orientation and transformations to map these into a global 3D structure. You’ll gain a deeper understanding of why quaternions work better than rotation ...
AlphaFold Decoded: Series Introduction
Просмотров 1,2 тыс.3 месяца назад
Welcome to the journey of “AlphaFold Decoded”! In this introduction, we’ll dive into what makes this series truly special. It’s more than just a deep dive into tensors, machine learning, and AlphaFold’s cutting-edge techniques - it’s an invitation to explore the incredible potential of computational biology together. Whether you’re here to expand your skills or simply curious about the magic be...
AlphaFold Decoded: Feature Embedding (Lesson 6)
Просмотров 2883 месяца назад
In the sixth video of this series, we explore Feature Embedding, a key aspect of AlphaFold’s input processing. You’ll delve into the workings of AlphaFold’s input embedder, the recycling embedder, and the ExtraMSA Stack. These modules are essential for preparing and refining the input data before it reaches the Evoformer, ensuring that the model has a comprehensive understanding of the protein’...
AlphaFold Decoded: Evoformer (Lesson 5)
Просмотров 6793 месяца назад
In the fifth video of this series, we dive into the Evoformer, a critical component of AlphaFold’s architecture. You’ll learn how the Evoformer processes and refines protein sequence and structure information through a series of innovative blocks. This deep dive will show you how AlphaFold builds complex relationships between residues to achieve high-precision protein structure predictions. In ...
AlphaFold Decoded: Feature Extraction (Lesson 4)
Просмотров 7244 месяца назад
In the fourth video of this series, we take an in-depth look at feature extraction in AlphaFold. You’ll see how to process the foundational data AlphaFold starts with, converting it into tensors for machine learning use. This video covers the entire feature extraction process, essential for building a full, workable implementation of AlphaFold. In the Jupyter Notebook, you’ll follow detailed st...
AlphaFold Decoded: Attention (Lesson 3)
Просмотров 6564 месяца назад
For the third video in this series, we take a look at attention. It's the core mechanism that drives most of what's shaping machine learning at the moment. This video is quite short, but pretty much complete regarding what you should know about attention. In the Jupyter Notebook for this video, you'll implement the MultiHeadAttention module that we'll use in AlphaFold. Aside of that, the notebo...
AlphaFold Decoded: Introduction to Machine Learning (Lesson 2)
Просмотров 9205 месяцев назад
This video is an introduction to Machine Learning, going through the example of a feed-forward neural network for handwritten digit recognition. You can explore the topic hands-on in the accompanying Jupyter Notebook, where you'll do the full implementation yourself with automatic checks: Assignment: 1) Clone github.com/kilianmandon/alphafold-decoded 2) Work on tutorials/machine_learning Setup ...
AlphaFold Decoded: Introduction to Tensors (Lesson 1)
Просмотров 3,4 тыс.6 месяцев назад
This introductory video is the perfect starting point for understanding the innovative protein structure modeling software, AlphaFold. Mastering tensors will be essential as you progress. For more details and to access the corresponding Jupyter notebook, visit our website: Assignment: 1) Clone github.com/kilianmandon/alphafold-decoded 2) Work on tutorials/tensor_introduction Setup Instructions ...
Great videos but your eyes seem red. Hope youre getting enough sleep my friend!! Really appreciate your work here.
hello, thanks for the great video! I found a little bug while studying. the $\bold{n}$ doesn't work in the markdown cell for the geometry.ipynb file in google colab. $\mathbf{n}$ works btw
Thanks for such a great video! I'm writing this just in case if there are other people who were stuck as I was. It took me a while to understand why you were adding up the probabilities during cluster masking. The point where i misunderstood was that we are trying to mask each *amino acid* which is represented as a row of one hot vectors (1,23) and not each elements of the matrix (which would just be 0 or 1). Therefore we add up all the probabilites for each row (which corresponds to one amino acid) and use torch.distributions.Categorical to pick just one in that row. And later change that value if it happens to be masked(which is 15% of every amino acids)
Keep up the good work
He is staring into my soul.
great tutorial ! Thank you very much for the great work ! salute.
You're a fantastic communicator, thank you. Would love to chat.
are you an AI?
great video, keep up the good work
Thanks, will do!
Wonderful project. Wondering if you're going to Cover AF3?
I'd love too, all molecules models are so interesting. But I won't be able to before they do code release because I can't check my modules, and it will be more difficult because they don't do PyTorch. This series worked so well because OpenFold was already a PyTorch implementation that I could use to compare the results. But fingers crossed 🤞🏻
Nice, dude! And I don't care what they say about you recording this while being held hostage at gunpoint by DeepMind, with blood saturated with Adderall, I believe you're just really passionate about this stuff, and I'm gonna binge watch all your videos!
Haha, if you thought that intro was intense, just wait-the first few episodes were definitely a challenge to film. But I’m really excited that you’re diving in, let me know how it goes!
What a time to be alive!
Haha lol
Thank you so much for this video
Most welcome 😊
Great intro! Your personal journey really inspires me to dive deeper into ML. As a life sciences student, your explanations make it clear how relevant this field is.
Wow, thank you!
Thank you so much for this amazing series! I’ve found the in-depth exploration of AlphaFold’s theory and implementation incredibly helpful. I’m curious to know if you have any plans to create a series on AlphaFold 3 in the future? I’m really interested in this field and would love to see more content on the latest developments. Thanks for all your hard work!
Thank you so much for the support! I’m really glad the series has been helpful for you. I do plan to dive into AlphaFold 3, but it might be a bit more challenging. For AlphaFold 2, I leaned on OpenFold as a working PyTorch implementation to validate intermediate results. With AlphaFold 3, I'll need to spend some time learning about diffusion models first. That said, I’m definitely planning to get into it after the code release-especially since the all-atom models are super exciting to me too. Thanks for your patience!
I'm also waiting for the AlphaFold 3 videos! However, note that according to Google's post the official code won't be released. Nonetheless, you can fond the pseudo code for it, and there are some unofficial open source versions you can check.
Thank you for your help!
You're welcome!
That's a very comprehensive summary for the input feature of AlphaFold2, which helps clarify things a lot. Thanks very much for your work!
Glad you enjoyed it!
Very good video again :) can’t wait for the next one
Thank you so much for the video it was so interesting!
Love it! Great video!
🤩
Super Video !
this is wonderful. thank you brother.
I really enjoy this Tutorial. I am looking forward to see the AlphaFold specific lessons and really learn how the mystery of AlphaFold unravels. And thanks for the recommendation of other good tutorials on machine learning ❤
Wow this video made my day! Thank you!
Hey everyone, at 13:21 the denominators for the second and third row should be e^3+e^4+e^5 and e^6+e^7+e^8 respectively, instead of e^0+e^1+e^2. Thanks for letting me know in the comments!
13:25 I could be mistaken here, but are the denominators in the second and third rows wrong? shouldn't it be e^3+e^4+e^5 and e^6+e^7+e^8 respectively instead of e^1+e^2+e^3, assuming you are doing Softmax(dim=1)
You're absolutely right, thanks for pointing that out! I'll start a pinned comment for errors :)
Please finish this series - looking forward to the whole thing. Breaking down AlphaFold is going to have huge value. I have one suggestion. While you can, start off the series differently with the high-level on AlphaFold first/videos on that and a more first principled breakdown of what we're building/high-level understanding before going into the code. Right now, the video about tensors without context on much else is less interesting. Especially because things that lack context don't tap into the right motivation circuits to make people interested, whereas if you explain first, then talk about the topics, it will be far more engaging. Good luck with this channel!
Thank you so much for your feedback and encouragement! I really appreciate your suggestion and completely agree. Starting with a high-level overview of AlphaFold before diving into the code would definitely make the series more engaging and informative. I as well have found that providing context first helps people stay motivated through the detailed parts. The video on tensors was indeed a bit heavy on syntax, which can be less interesting without a bigger picture. I'll definitely incorporate a more principled breakdown in future videos. While I'll certainly try to put everything into a bigger picture in the context of AlphaFold, the first three videos I have planned will serve as a general introduction to machine learning concepts. These will touch on AlphaFold but primarily use more classic examples like images and NLP to explain the mechanisms, as these topics are often easier to interpret and will set a strong foundation. Thank you again for the tip, and I'm excited to continue this journey with you all. Stay tuned!
Can't wait for more lessons.
Really appreciate your kind words!
Wow! I saw this video and instantly subscribed! That is great value content. Thanks for sharing. Greetings from Brazil!
I really appreciate your kind comment! More videos are on the way. Greetings back from Germany 👋
Great job!! I could follow along really well so far and I‘m already looking forward to the next episode!
Your support is amazing, thank you. I hope you enjoy what's next!
A very helpful tutorial, thank you very much
Thanks for you kind words!
This is amazing!
Your feedback is appreciated; exciting videos coming up!
nice and simple explanation
Amazing, thank you for your nice feedback! I look forward to your opinion on the next chapters.
Thanks for explaining it so well :) So excited to learn more in the next video!!
Thanks so much for watching and your support!
I’ve been waiting for a tutorial like this for so long! Can’t wait to learn more about AlphaFold!!
I am super glad that this tutorial was helpful for you. Please stayed tuned for the upcoming videos and let us know what you think!
Wow thank you so much for this tutorial
Thanks for the comment and for watching!
What a great and concise tutorial, will definitely follow along once I have time😊
Thanks a lot; can't wait for you to see the next video!