Hot Topics #15 (Jan. 23, 2023)
Generative models of SARS-CoV-2 and antibodies, open-ended RL, TD-learning, GNNs, and more.
A deep generative model of the SARS-CoV-2 spike protein predicts future variants: Rahul M. Dhodapkar; Jan. 18, 2023
Abstract: SARS-CoV-2 has demonstrated a robust ability to adapt in response to environmental pressures—increasing viral transmission and evading immune surveillance by mutating its molecular machinery. While viral sequencing has allowed for the early detection of emerging variants, methods to predict mutations before they occur remain limited. This work presents SpikeGPT2, a deep generative model based on ProtGPT2 and fine-tuned on SARS-CoV-2 spike (S) protein sequences deposited in the NIH Data Hub before May 2021. SpikeGPT2 achieved 88.8% next-residue prediction accuracy and successfully predicted amino acid substitutions found only in a held-out set of spike sequences deposited on or after May 2021, to which SpikeGPT2 was never exposed. When compared to several other methods, SpikeGPT2 achieved the best performance in predicting such future mutations. SpikeGPT2 also predicted several novel variants not present in the NIH SARS-CoV-2 Data Hub. A binding affinity analysis of all 54 generated substitutions identified 5 (N439A, N440G, K458T, L492I, and N501Y) as predicted to simultaneously increase S/ACE2 affinity, and decrease S/tixagevimab+cilgavimab affinity. Of these, N501Y has already been well-described to increase transmissibility of SARS-CoV-2. These findings indicate that SpikeGPT2 and other similar models may be employed to identify high-risk future variants before viral spread has occurred.
Human-Timescale Adaptation in an Open-Ended Task Space: DeepMind Adaptive Agent Team; Jan. 18, 2023
Abstract: Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement learning (RL). In this work, we demonstrate that training an RL agent at scale leads to a general in-context learning algorithm that can adapt to open-ended novel embodied 3D problems as quickly as humans. In a vast space of held-out environment dynamics, our adaptive agent (AdA) displays on-the-fly hypothesis-driven exploration, efficient exploitation of acquired knowledge, and can successfully be prompted with first-person demonstrations. Adaptation emerges from three ingredients: (1) meta-reinforcement learning across a vast, smooth and diverse task distribution, (2) a policy parameterised as a large-scale attention-based memory architecture, and (3) an effective automated curriculum that prioritises tasks at the frontier of an agent's capabilities. We demonstrate characteristic scaling laws with respect to network size, memory length, and richness of the training task distribution. We believe our results lay the foundation for increasingly general and adaptive RL agents that perform well across ever-larger open-ended domains.
Everything is Connected: Graph Neural Networks: Petar Veličković; Jan. 19, 2023
Abstract: In many ways, graphs are the main modality of data we receive from nature. This is due to the fact that most of the patterns we see, both in natural and artificial systems, are elegantly representable using the language of graph structures. Prominent examples include molecules (represented as graphs of atoms and bonds), social networks and transportation networks. This potential has already been seen by key scientific and industrial groups, with already-impacted application areas including traffic forecasting, drug discovery, social network analysis and recommender systems. Further, some of the most successful domains of application for machine learning in previous years -- images, text and speech processing -- can be seen as special cases of graph representation learning, and consequently there has been significant exchange of information between these areas. The main aim of this short survey is to enable the reader to assimilate the key concepts in the area, and position graph representation learning in a proper context with related fields.
An Analysis of Quantile Temporal-Difference Learning: Rowland et al.; Jan, 11, 2023
Abstract: We analyse quantile temporal-difference learning (QTD), a distributional reinforcement learning algorithm that has proven to be a key component in several successful large-scale applications of reinforcement learning. Despite these empirical successes, a theoretical understanding of QTD has proven elusive until now. Unlike classical TD learning, which can be analysed with standard stochastic approximation tools, QTD updates do not approximate contraction mappings, are highly non-linear, and may have multiple fixed points. The core result of this paper is a proof of convergence to the fixed points of a related family of dynamic programming procedures with probability 1, putting QTD on firm theoretical footing. The proof establishes connections between QTD and non-linear differential inclusions through stochastic approximation theory and non-smooth analysis.
Unlocking de novo antibody design with generative artificial intelligence: Shanehsazzadeh et al.; Jan. 9, 2023
Abstract: Generative artificial intelligence (AI) has the potential to greatly increase the speed, quality and controllability of antibody design. Traditional de novo antibody discovery requires time and resource intensive screening of large immune or synthetic libraries. These methods also offer little control over the output sequences, which can result in lead candidates with sub-optimal binding and poor developability attributes. Several groups have introduced models for generative antibody design with promising in silico evidence [1–10], however, no such method has demonstrated de novo antibody design with experimental validation. Here we use generative deep learning models to de novo design antibodies against three distinct targets, in a zero-shot fashion, where all designs are the result of a single round of model generations with no follow-up optimization. In particular, we screen over 400,000 antibody variants designed for binding to human epidermal growth factor receptor 2 (HER2) using our high-throughput wet lab capabilities. From these screens, we further characterize 421 binders using surface plasmon resonance (SPR), finding three that bind tighter than the therapeutic antibody trastuzumab. The binders are highly diverse, have low sequence identity to known antibodies, and adopt variable structural conformations. Additionally, these binders score highly on our previously introduced Naturalness metric [11], indicating they are likely to possess desirable developability profiles and low immunogenicity. We open source1 the HER2 binders and report the measured binding affinities. These results unlock a path to accelerated drug creation for novel therapeutic targets using generative AI combined with high-throughput experimentation.
Evolutionary selection of proteins with two folds; Schafer and Porter; Jan. 20, 2023
Abstract: Although most globular proteins fold into a single stable structure, an increasing number have been shown to remodel their secondary and tertiary structures in response to cellular stimuli. State-of-the-art algorithms predict that these fold-switching proteins assume only one stable structure, missing their functionally critical alternative folds. Why these algorithms predict a single fold is unclear, but all of them infer protein structure from coevolved amino acid pairs. Here, we hypothesize that coevolutionary signatures are being missed. Suspecting that over-represented single-fold sequences may be masking these signatures, we developed an approach to search both highly diverse protein superfamilies--composed of single-fold and fold-switching variant--and protein subfamilies with more fold-switching variants. This approach successfully revealed coevolution of amino acid pairs uniquely corresponding to both conformations of 56/58 fold-switching proteins from distinct families. Then, using a set of coevolved amino acid pairs predicted by our approach, we successfully biased AlphaFold2 to predict two experimentally consistent conformations of a candidate protein with unsolved structure. The discovery of widespread dual-fold coevolution indicates that fold-switching sequences have been preserved by natural selection, implying that their functionalities provide evolutionary advantage and paving the way for predictions of diverse protein structures from single sequences.
Official code for "Unlocking de novo antibody design"; Sequences of binding antibodies and corresponding affinity values identified in the study Unlocking de novo antibody design with generative artificial intelligence.
We present data for zero-shot HCDR3 designs (zero-shot-designs.csv) and multi-step multi-HCDR designs (multi-step-binders.csv).
TorchMD; TorchMD intends to provide a simple to use API for performing molecular dynamics using PyTorch. This enables researchers to more rapidly do research in force-field development as well as integrate seamlessly neural network potentials into the dynamics, with the simplicity and power of PyTorch.
LangChain; Large language models (LLMs) are emerging as a transformative technology, enabling developers to build applications that they previously could not. But using these LLMs in isolation is often not enough to create a truly powerful app - the real power comes when you can combine them with other sources of computation or knowledge.
This library is aimed at assisting in the development of those types of applications.
Image Mixer Demo; Provide one or more images to be mixed together by a fine-tuned Stable Diffusion model.
