Recently, utilizing reinforcement learning (RL) to generate molecules with desired properties has been highlighted as a promising strategy for drug design. A molecular docking program - a physical simulation that estimates protein-small molecule binding affinity - can be an ideal reward scoring function for RL, as it is a straightforward proxy of the therapeutic potential. Still, two imminent challenges exist for this task. First, the models often fail to generate chemically realistic and pharmacochemically acceptable molecules. Second, the docking score optimization is a difficult exploration problem that involves many local optima and less smooth surfaces with respect to molecular structure. To tackle these challenges, we propose a novel RL framework that generates pharmacochemically acceptable molecules with large docking scores. Our method - Fragment-based generative RL with Explorative Experience replay for Drug design (FREED) - constrains the generated molecules to a realistic and qualified chemical space and effectively explores the space to find drugs by coupling our fragment-based generation method and a novel error-prioritized experience replay (PER). We also show that our model performs well on both de novo and scaffold-based schemes. Our model produces molecules of higher quality compared to existing methods while achieving state-of-the-art performance on two of three targets in terms of the docking scores of the generated molecules. We further show with ablation studies that our method, predictive error-PER (FREED(PE)), significantly improves the model performance.

This article is an on-site version of the #techAsia newsletter.sign up here Send newsletter directly to your inbox every Wednesday Hello, Kenji from Tokyo this week is currently undergoing home quarantine for Covid-19. For our big story, there is another scoop about Apple from Nikkei Asia. China's state-owned enterprise has become a supplier of the latest flagship iPhone displays. This shows how advanced China's technology, including artificial intelligence, has advanced, as warned by a former Pentagon chief software officer (Mercedes Top 10). Meanwhile, China is building and diversifying its sources of strategic mineral resources, including lithium, a key component of the world's leading electric vehicle industry (our views, smart data and spotlights).

Clustering time series data before fitting can improve accuracy by 33% -- src. In 2021, researchers at UCLA developed a method that can improve model fit on many different time series'. By aggregating similarly structured data and fitting a model to each group, our models can specialize. While fairly straightforward to implement, as with any other complex deep learning method, we are often computationally limited by large data sets. However, all of the methods listed have support in both R and python, so development on smaller datasets should be pretty "simple."

Chatbots have a checkered past of often not delivering the performance their providers have promised. This is especially true in the IT service management (ITSM) and multilingual NLP spaces, where service desks found support teams deluged with complaints -- yes, about the support chatbots. Just getting English language nuance right and how enterprises communicate often require chatbots to be custom programmed with constraint and logic workflows supported with natural language processing (NLP) and machine learning. If that sounds like a science project, it is, and IT users are the test subjects. Because of their complexity, chatbots were contributing to already overflowing trouble-ticket queues.

In this community review report, we discuss applications and techniques for fast machine learning (ML) in science -- the concept of integrating power ML methods into the real-time experimental data processing loop to accelerate scientific discovery. The material for the report builds on two workshops held by the Fast ML for Science community and covers three main areas: applications for fast ML across a number of scientific domains; techniques for training and implementing performant and resource-efficient ML algorithms; and computing architectures, platforms, and technologies for deploying these algorithms. We also present overlapping challenges across the multiple scientific domains where common solutions can be found. This community report is intended to give plenty of examples and inspiration for scientific discovery through integrated and accelerated ML solutions. This is followed by a high-level overview and organization of technical advances, including an abundance of pointers to source material, which can enable these breakthroughs.

The MXC Foundation has made a remarkable entry into the nascent multi-billion dollar AI smart device market. With the exponential growth of its network across the globe, the Foundation is thrilled to introduce more aspects to its network usage, allowing its mining community to utilize the data republic and see the network in action. The proprietary MXProtocol, together with scalable and secure aspects of device provisioning that connect with sensor technology, has proven successful and brings us a step closer to realizing truly smart cities. One such use case, which the MXC Foundation recently tested in a controlled environment, was the Edge X AI Camera. Read on to find out more about all the great functionalities packed into one small device.

A partnership between the Massachusetts Institute of Technology and the chemical giant BASF has managed to successfully create an AI-driven process to speed up the discovery of custom 3D printing materials. Chemists usually develop a few iterations of a material candidate over a couple of days and test them in the lab. The new machine-learning algorithm can churn out hundreds of those iterations with the desired characteristics in the same timeframe. This would save time and raw material costs, as well as lessen the environmental impact of the discarded chemicals. Not only that, but the algorithm may also come up with ideas that the material's engineer could have overlooked for various reasons.

Researchers at MIT and BASF have developed a data-driven system that accelerates the process of discovering new 3D printing materials that have multiple mechanical properties. A new machine-learning system costs less, generates less waste, and can be more innovative than manual discovery methods. The growing popularity of 3D printing for manufacturing all sorts of items, from customized medical devices to affordable homes, has created more demand for new 3D printing materials designed for very specific uses. To cut down on the time it takes to discover these new materials, researchers at MIT have developed a data-driven process that uses machine learning to optimize new 3D printing materials with multiple characteristics, like toughness and compression strength. By streamlining materials development, the system lowers costs and lessens the environmental impact by reducing the amount of chemical waste.

A team of researchers at Cornell Tech, Cornell University's tech-focused research campus, has developed a forecast for how technologies like artificial intelligence could shape cities in the coming decade. After a year of work, the team released its first "Horizon Scan" report last week to discuss the potential risks and applications of recent advancements in urban tech. The forecast report predicts areas where the most radical and rapid changes in urban tech could take place, touching on topics such as "supercharged" smart city infrastructure, the use of sustainable building materials and machine learning in the public sector, among other areas of interest. The project was led by Anthony Townsend, urbanist in residence at the Jacobs Urban Tech Hub at Cornell Tech, who has spent years studying tech-related issues like the digital divide. He said the goal of the Horizon Scan was to create a road map "to make better decisions about applied research" in urban tech. Townsend said the need to weigh potential pros and cons of machine learning's applications in the public sector is a recurring factor in the report.

Self-supervised learning holds promise to revolutionize molecule property prediction - a central task to drug discovery and many more industries - by enabling data efficient learning from scarce experimental data. Despite significant progress, non-pretrained methods can be still competitive in certain settings. We reason that architecture might be a key bottleneck. In particular, enriching the backbone architecture with domain-specific inductive biases has been key for the success of self-supervised learning in other domains. In this spirit, we methodologically explore the design space of the self-attention mechanism tailored to molecular data. We identify a novel variant of self-attention adapted to processing molecules, inspired by the relative self-attention layer, which involves fusing embedded graph and distance relationships between atoms. Our main contribution is Relative Molecule Attention Transformer (R-MAT): a novel Transformer-based model based on the developed self-attention layer that achieves state-of-the-art or very competitive results across a~wide range of molecule property prediction tasks.