Goals

• Learn about machine learning methods applied to problems in structural biology
• Learn how to critically read and evaluate papers
• Learn how to pose research problems and practice oral and written scientific communication skills
• Bonus: Exposure to relevant basic and applied ML research in industry from guest speakers

Topics

A non-exhaustive list of topics we will cover include:

• An introduction to structural biology
• Protein structure prediction before and after AlphaFold2
• Computer vision and cryo-electron microscopy (cryo-EM)
• Computational protein design, in particular, antibody and vaccine design
• Physics-based modeling and statistical mechanics
• Small molecule drug discovery

Selected papers will cover a broad range of algorithmic concepts and machine learning techniques including:

• Supervised learning and designing appropriate benchmarks and metrics
• Language modeling and transformers
• Generative modeling techniques including VAEs, GANs, normalizing flows, and diffusion models
• Geometric deep learning
• Neural fields and multi-view 3D reconstruction

In addition to the assigned papers, optional primers or reviews on relevant topics will be made available for background reading.

Assignments

Flash talk assignment info: link

Final project guidelines: link

Guest Speakers

Monday, September 25th, 4:30pm ET
John Jumper (DeepMind)

Title: Highly accurate protein structure prediction with deep learning

Abstract: Our work on deep learning for biology, specifically the AlphaFold system, has demonstrated that neural networks are capable of highly accurate modeling of both protein structure and protein-protein interactions. In particular, the system shows a remarkable ability to extract chemical and evolutionary principles from experimental structural data. This computational tool has repeatedly shown the ability to not only predict accurate structures for novel sequences and novel folds but also to do unexpected tasks such as selecting stable protein designs or detecting protein disorder. In this lecture, I will discuss the context of this breakthrough in the machine learning principles, the diverse data and rigorous evaluation environment that enabled it to occur, and the many innovative ways in which the community is using these tools to do new types of science. I will also reflect on some surprising limitations -- insensitivity to mutations and the lack of context about the chemical environment of the proteins -- and how this may be traced back to the essential features of the training process. Finally, I will conclude with a discussion of some ideas on the future of machine learning in structure biology and how the experimental and computational communities can think about organizing their research and data to enable many more such breakthroughs in the future.

Bio: John Jumper received his PhD in Chemistry from the University of Chicago, where he developed machine learning methods to simulate protein dynamics. Prior to that, he worked at D.E. Shaw Research on molecular dynamics simulations of protein dynamics and supercooled liquids. He also holds an MPhil in Physics from the University of Cambridge and a B.S. in Physics and Mathematics from Vanderbilt University. At DeepMind, John is leading the development of new methods to apply machine learning to protein biology.

Thursday, November 16th, 3:00pm ET
Jason Yim (MIT)

Title: Diffusion models for protein structure and de novo design.

Abstract: Generative machine learning is revolutionizing protein design. In this talk, I will discuss recent advances in using diffusion models to generate protein structures and perform conditional generation towards protein design desiderata. First, I will go over FrameDiff, including an overview of the mathematical foundation of SE(3) diffusion and a practical algorithm for training a frame-based generative model over protein backbones. Next, I will overview how SE(3) diffusion is used in RFDiffusion, a state-of-the-art protein design method, which is pre-trained on protein structure prediction. We show that a single method, RFdiffusion, enables binder design, motif-scaffolding, and symmetric protein generation. Finally, I discuss current limitations and the technical challenges on the horizon for de novo protein design.

Bio: Jason Yim is a PhD candidate at the Massachusetts Institute of Technology (MIT) Computer Science and Artificial Intelligence Laboratory advised by Tommi Jaakkola and Regina Barzilay. His research focuses on developing generative models for scientific applications as well as experimental design in biological experiments. He has previously worked as a research engineer at DeepMind and interned at Microsoft AI4science.

Thursday, December 7th, 3:00pm ET
Stephan Eismann (AtomicAI)

Title: Enabling structure-based drug discovery for RNA using AI

Abstract: RNA molecules adopt three-dimensional structures that are critical to their function and of interest in drug discovery. Few RNA structures are known, however, and predicting them computationally has proven challenging. I will talk about ARES, a machine learning approach that enables identification of accurate structural models without assumptions about their defining characteristics, despite being trained with only 18 known RNA structures. ARES outperforms previous methods and consistently produced the best results in community-wide blind RNA structure prediction challenges. In addition to ARES, I will talk about recent advancements in tertiary RNA structure prediction at Atomic AI.

Bio: Stephan leads the ML team at Atomic AI. Originally from Germany, he did his PhD at Stanford University where his research focused on the development of novel ML algorithms for problems in structural biology