Hot Topics #20 (Mar. 1, 2023)
Protein design, robot learning, and imitation learning.
De novo design of luciferases using deep learning: Yeh et al., Feb. 22, 2023
Abstract: De novo enzyme design has sought to introduce active sites and substrate-binding pockets that are predicted to catalyse a reaction of interest into geometrically compatible native scaffolds1,2, but has been limited by a lack of suitable protein structures and the complexity of native protein sequence–structure relationships. Here we describe a deep-learning-based ‘family-wide hallucination’ approach that generates large numbers of idealized protein structures containing diverse pocket shapes and designed sequences that encode them. We use these scaffolds to design artificial luciferases that selectively catalyse the oxidative chemiluminescence of the synthetic luciferin substrates diphenylterazine3 and 2-deoxycoelenterazine. The designed active sites position an arginine guanidinium group adjacent to an anion that develops during the reaction in a binding pocket with high shape complementarity. For both luciferin substrates, we obtain designed luciferases with high selectivity; the most active of these is a small (13.9 kDa) and thermostable (with a melting temperature higher than 95 °C) enzyme that has a catalytic efficiency on diphenylterazine (kcat/Km = 106 M−1 s−1) comparable to that of native luciferases, but a much higher substrate specificity. The creation of highly active and specific biocatalysts from scratch with broad applications in biomedicine is a key milestone for computational enzyme design, and our approach should enable generation of a wide range of luciferases and other enzymes.
Paper URL: https://www.nature.com/articles/s41586-023-05696-3
Efficient and scalable de novo protein design using a relaxed sequence space: Frank et al., Feb. 25, 2023
Abstract: Deep learning techniques are being used to design new proteins by creating target backbone geometries and finding sequences that can fold into those shapes. While methods like ProteinMPNN provide an efficient algorithm for generating sequences for a given protein backbone, there is still room for improving the scope and computational efficiency of backbone generation. Here, we report a backbone hallucination protocol that uses a relaxed sequence representation. Our method enables protein backbone generation using a gradient descent driven hallucination approach and offers orders-of-magnitude efficiency enhancements over previous hallucination approaches. We designed and experimentally produced over 50 proteins, most of which expressed well in E. Coli, were soluble and adopted the desired oligomeric state along with the correct composition of secondary structure as measured by CD. Exemplarily, we determined 3D electron density maps using single-particle cryo EM analysis for three single-chain de-novo proteins comprising 600 AA which closely matched with the designed shape. These have no structural analogues in the protein data bank (PDB), representing potentially novel folds or arrangement of domains. Our approach broadens the scope of de novo protein design and contributes to accessibility to a wider community.
Paper URL: https://www.biorxiv.org/content/10.1101/2023.02.24.529906v1
Cyclic peptide structure prediction and design using AlphaFold: Rettie et al., Feb. 26, 2023
Abstract: Deep learning networks offer considerable opportunities for accurate structure prediction and design of biomolecules. While cyclic peptides have gained significant traction as a therapeutic modality, developing deep learning methods for designing such peptides has been slow, mostly due to the small number of available structures for molecules in this size range. Here, we report approaches to modify the AlphaFold network for accurate structure prediction and design of cyclic peptides. Our results show this approach can accurately predict the structures of native cyclic peptides from a single sequence, with 36 out of 49 cases predicted with high confidence (pLDDT > 0.85) matching the native structure with root mean squared deviation (RMSD) less than 1.5 Å. Further extending our approach, we describe computational methods for designing sequences of peptide backbones generated by other backbone sampling methods and for de novo design of new macrocyclic peptides. We extensively sampled the structural diversity of cyclic peptides between 7–13 amino acids, and identified around 10,000 unique design candidates predicted to fold into the designed structures with high confidence. X-ray crystal structures for seven sequences with diverse sizes and structures designed by our approach match very closely with the design models (root mean squared deviation < 1.0 Å), highlighting the atomic level accuracy in our approach. The computational methods and scaffolds developed here provide the basis for custom-designing peptides for targeted therapeutic applications.
Paper URL: https://www.biorxiv.org/content/10.1101/2023.02.25.529956v1
NeRF in the Palm of Your Hand: Corrective Augmentation for Robotics via Novel-View Synthesis: Zhou et al., Jan. 18, 2023
Abstract: Expert demonstrations are a rich source of supervision for training visual robotic manipulation policies, but imitation learning methods often require either a large number of demonstrations or expensive online expert supervision to learn reactive closed-loop behaviors. In this work, we introduce SPARTN (Synthetic Perturbations for Augmenting Robot Trajectories via NeRF): a fully-offline data augmentation scheme for improving robot policies that use eye-in-hand cameras. Our approach leverages neural radiance fields (NeRFs) to synthetically inject corrective noise into visual demonstrations, using NeRFs to generate perturbed viewpoints while simultaneously calculating the corrective actions. This requires no additional expert supervision or environment interaction, and distills the geometric information in NeRFs into a real-time reactive RGB-only policy. In a simulated 6-DoF visual grasping benchmark, SPARTN improves success rates by 2.8× over imitation learning without the corrective augmentations and even outperforms some methods that use online supervision. It additionally closes the gap between RGB-only and RGB-D success rates, eliminating the previous need for depth sensors. In real-world 6-DoF robotic grasping experiments from limited human demonstrations, our method improves absolute success rates by 22.5% on average, including objects that are traditionally challenging for depth-based methods. See video results at \url{https://bland.website/spartn}
Paper URL: https://arxiv.org/abs/2301.08556
CACTI: A Framework for Scalable Multi-Task Multi-Scene Visual Imitation Learning: Mandi et al., Dec. 12, 2022
Abstract: Large-scale training have propelled significant progress in various sub-fields of AI such as computer vision and natural language processing. However, building robot learning systems at a comparable scale remains challenging. To develop robots that can perform a wide range of skills and adapt to new scenarios, efficient methods for collecting vast and diverse amounts of data on physical robot systems are required, as well as the capability to train high-capacity policies using such datasets. In this work, we propose a framework for scaling robot learning, with specific focus on multi-task and multi-scene manipulation in kitchen environments, both in simulation and in the real world. Our proposed framework, CACTI, comprises four stages that separately handle: data collection, data augmentation, visual representation learning, and imitation policy training, to enable scalability in robot learning . We make use of state-of-the-art generative models as part of the data augmentation stage, and use pre-trained out-of-domain visual representations to improve training efficiency. Experimental results demonstrate the effectiveness of our approach. On a real robot setup, CACTI enables efficient training of a single policy that can perform 10 manipulation tasks involving kitchen objects, and is robust to varying layouts of distractors. In a simulated kitchen environment, CACTI trains a single policy to perform 18 semantic tasks across 100 layout variations for each individual task. We will release the simulation task benchmark and augmented datasets in both real and simulated environments to facilitate future research.
Paper URL: https://arxiv.org/abs/2212.05711
alphafold_finetune; Python code for fine-tuning AlphaFold to perform protein-peptide binding predictions. This repository is a collaborative effort: Justas Dauparas implemented the AlphaFold changes necessary for fine-tuning and wrote a template of the fine-tuning script. Amir Motmaen and Phil Bradley further developed and extensively tested the fine-tuning and inference scripts in the context of protein-peptide binding.
Quantum deep field for molecule; This repository provides a simple implementation of the quantum deep field (QDF) framework for molecules proposed in our study as follows.
Masashi Tsubaki and Teruyasu Mizoguchi
Quantum deep field: data-driven wave function, electron density generation, and energy prediction and extrapolation with machine learning
Physical Review Letters, 2020
https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.125.206401Masashi Tsubaki and Teruyasu Mizoguchi
On the equivalence of molecular graph convolution and molecular wave function with poor basis set
Advances in Neural Information Processing Systems, 2020
https://proceedings.neurips.cc/paper/2020/hash/1534b76d325a8f591b52d302e7181331-Abstract.html
QDF is a machine learning model that provides the electron density ρ of molecules by learning the atomization energy E of molecules on a large dataset (e.g., the QM9 dataset [1]). The QDF model involves a linear component (i.e., the linear combination of atomic orbitals, LCAO [2]) and two nonlinear components (i.e., the energy functional and the Hohenberg-Kohn map [3]), in which the latter two are implemented by deep neural network (DNN) (see the above figure). In particular, the DNN-based Hohenberg-Kohn map serves as a physical, external potential constraint on ψ (i.e., the Kohn-Sham molecular orbitals) in learning the energy functional E = F[ψ] based on the density functional theory. For more details read our papers, in which the NeurIPS paper provides the equivalence and difference between LCAO and graph neural networks or graph convolutional networks for molecules.