2019 SEMINAR ARCHIVE
January 9th 2019
speaker: Krishna Praneeth Kilambi
title: Integrating structural data with the human interactome to support target discovery and characterization
abstract: Talk overview:
Introduce the human protein–protein interaction (PPI) network as a framework to better understand therapeutic targets
Combine structural information with the human interactome to enhance the network
Interpret the effects of disease-associated mutations on PPI to allow comparison of normal and disease states
speaker: Andreas Lehmann
title: Model of a solubilizing GDF11 prodomain peptide as an example for protein modeling & design at Biogen
abstract: Growth differentiation factor 11 (GDF11) is a member of the transforming growth factor β (TGF-β) family. Members of this protein family play very diverse roles in mammalian development and homeostasis, and have tremendous therapeutic potential. I will discuss a semi-artisanal model (borrowing Chris’ poetic description of this) I made to illustrate the potential conformation of a peptide that can be proteolytically cleaved from GDF11’s prodomain. This prodomain peptide keeps the otherwise insoluble homodimeric GDF11 soluble and does not interfere with its activity. Using this example, I aim to illustrate the use of protein modeling & design as an embedded function of biologics drug discovery R&D at Biogen, the oldest independent biotech company in the world.
February 6th 2019
speaker: Samaneh Mesbahi-Vasey
title: how to build entirely novel, functional proteins de novo with SEWING
abstract: De novo protein design traditionally has begun with the researchers specifying a protein structure to be created and then using molecular modeling to identify amino acid sequences that will adopt that fold. However, when designing for protein function, we no longer need structure-driven design. I will present a method called SEWING which allows design for protein function without the need to know/predict the desired protein’s structure a priori
speaker: Chris Bahl
title: the beast is slain; LayerDesign is dead but not gone (yet)
abstract: I'll review LayerDesign, why and where it's useful when performing design in Rosetta. Then, I'll walk through how to do LayerDesign without using LayerDesign; this relies on several different residue selectors and the new DesignRestrictions task operation. Finally, I'll discuss the new recommended changes to default LayerDesign behavior.
March 6th 2019
speaker: Shourya Sonkar Roy Burman
title: Modeling Interactions of Flexible Proteins
abstract: A story of two related methods:
RosettaDock 4.0: a heterodimer docking protocol that incorporates flexibility by efficiently simulating conformational selection from hundreds of pre-generated backbone conformations and identifies the near-native models with a novel coarse-grained score function called Motif Dock Score (MDS)
Rosetta SymDock2: a symmetric protein assembly method that leverages MDS in the coarse-grained phase and simulates subunit flexibility through induced fit by all-atom flexible-backbone refinement
April 3rd 2019
speaker: Surge Biswas
title: Unified rational protein engineering using sequence-only deep representation learning
abstract: We use deep representation learning to learn a semantically rich basis in which we can quantitatively compare proteins given only their amino acid sequences. This paradigm enables stability and quantitative function prediction and shows promise for protein engineering. I’ll talk about these results, and will discuss future directions related to structure prediction, data efficient protein design, and specific applications thereof that we’re exploring in the lab.
May 1st 2019
speaker: Sergey Ovchinnikov
title: A workshop on using sequence models for protein design (pssm, gremlin/mrf, vae, seq2seq autoencoders etc)
abstract: topics:
1) intro to google colab
2) intro to tensorflow and optimization
3) PSSM (simple one-body model)
4) Markov Random Fields (one-body, two-body model)
5) Autoencoders (simple mixture one-body model)
6) Variational Autoencoders
7) Sequence models
June 5th 2019
speaker: AJ Vincelli
title: Optimizing Rosetta’s Relax Algorithm to Prepare Published Protein X-Ray Structures
abstract: This project aims to improve the quality of X-ray crystal structures published in the PDB as an essential preparatory first step of biomedical and drug discovery initiatives. We have developed and calculated a six-factor aggregate “Quality Score” for 428 published X-ray structures, both before and after refinement with Rosetta's Relax algorithm, and found that the scores of low-quality structures improved significantly after Relaxation. Our next steps include expanding the dataset to protein:ligand complexes, and optimizing the refinement protocol for general use on protein X-ray structures found in the PDB.
speaker: Firas Khatib
title: Building Denovo Cryo-EM Structures Collaboratively with Video Gamers
abstract: With the rapid improvement of cryo-electron microscopy (Cryo-EM) resolution, new computational tools are needed to assist and improve upon atomic model building and refinement options. Microscopists can now collaborate with the players of the protein folding game Foldit to generate thorough high quality de novo structural models. This development could greatly speed the generation of excellent Cryo-EM structures when used in addition to current methods.
July 10th 2019
speaker: Per Greisen
title: Computational design of biosensors for small molecules
abstract: The ability to detect and respond to small molecules has great applications ranging from biotechnology to the pharmaceutical industry. Many natural biosensors are made from proteins and can function both inside cells/bacteria as well as outside of cell/bacteria. Unfortunately, there are not preexisting for all small molecules of interest but still a great need to generate biosensors. Computational protein design has shown its ability to generate new protein folds as well as create new functions in proteins. One way to generate new biosensors is to use computational protein design which in principle can design proteins binders with high affinity and specificity. Here, we will go through the successes and failures of computational protein design for small molecules with emphasis on Rosetta showing results from different small molecules (e.g. fentanyl, vitamin D/THC). Lastly, we will briefly discuss endeavours to solve and improve the protocols for computational design of small molecules.
August 14th 2019
speaker: Shourya Sonkar Roy Burman
title: Using Movers in PyRosetta
abstract: This will be a tutorial on how to use PyRosetta to write protocols. As an example, we will use PyRosetta to write code with basic movers. Next, we will explore the PyRosetta documentation to see what functions are available for given movers and write a more complicated protocol.
September 4th 2019
speaker: Nasos Dousis
title: Can we incorporate dynamics into routine enzyme design?
abstract: I will share some highlights from the Fall 2019 ACS meeting, and then guide a discussion on how we might incorporate dynamics to engineer catalytic rate in enzymes.
September 1st 2019
speaker: Alex Garruss
title: Deep Mutational Learning
abstract: I’ll present some machine learning approaches we are developing in the Church Lab to understand large-scale, deep mutational scanning experiments of protein function. We will discuss three areas of approach: sequence-only modeling (using embedding layers, local convolutional and graph convolutional neural networks), evolution-guided modeling (using latent variable models built from related sequences), and structural modeling (3D convolutions of all-atom structures) to predict the functional effect of varying a given protein’s sequence. Such approaches enable new understanding of function and are useful for objectives in protein engineering.
October 2nd 2019
speaker: Sergey Ovchinnikov
title: Improved protein structure prediction using predicted inter-residue orientations
abstract: I'll describe some improvements to our structure prediction protocol, more specifically using residual blocks to generate better restraints for rosetta modeling. (Aka, what we did to catch up to alphafold)
November 6th 2019
speaker: Victor Ovchinnikov
title: An Atomic Structure of the EmrE Multidrug Transporter for the Design of Inhibitor Peptides
abstract: Atomic-resolution structures of the EmrE multidrug transporter in ligand-free form, as well as bound to the ligands tetraphenylphosphonium+ (TPP) and ethidium+ are refined using multi-microsecond molecular dynamics simulations, using restraints to a low-resolution electron density map. We designed proteolysis-resistant stapled peptides with the aim of inhibiting the drug-resistance activity of EmrE by interfering with dimerization.
December 4th 2019
speaker: Frank Teets
title: Advances in Requirement-Driven Protein Design
abstract: Requirement-driven protein design is a historically underdeveloped approach to de novo design, relying on functional rather than structural constraints to guide the design process. Here, I present a series of computational improvements to the requirement-driven protein design algorithms in Rosetta that dramatically improve the flexibility and performance of the design pipeline.