IBM RoboRXN

2021: Ability to train RoboRXN based on native datasets

On December 1, 2021, IBM announced the expanded capabilities of RoboRXN, a cloud remote laboratory with artificial intelligence designed to help chemists quickly find and create new molecules and substances.

RoboRXN uses cloud technologies and artificial intelligence to optimize the discovery process. However, it is important that this process be safe for organizations working with proprietary data. There is also a growing need to increase the stability of production processes - this also applies to chemical reactions, during which raw materials are converted into finished products.

To solve these problems, IBM offered advanced RoboRXN capabilities:

• Security and Castomization - Cloud infrastructure capabilities enable users to train RXNs based on their own datasets, enabling safer experimentation and customization of prediction models using proprietary knowledge.
• More environmentally friendly chemical processes: the added AI models will help chemists quickly predict and identify more environmentally friendly enzymes (complex biomolecules necessary to obtain products such as paper, cosmetics, pharmaceuticals and flavors from materials).

The search and launch of one new material takes almost 10 years of work and from $10 million to $100 million. This is due to the fact that in chemical synthesis the trial and error method is still used and in fact steps are not taken to modernize the material development processes.

"To create a new material, chemists have to operate in an almost infinite space in which there are more potential chemical compounds than atoms in the universe," said Dr. Alessandro Curioni, PhD. Alessandro Curioni), Honorary Officer IBM and Director. IBM Research Europe- To help overcome challenges that require new materials, such as hunger, climate change, infections, researchers need to be able to effectively generate, synthesize, and test ideas for potential materials. Applying AI to this massive challenge through technologies such as RoboRXN has the potential to help increase the efficiency, sustainability and relevance of newly created materials to almost every industry. "

IBM RoboRXN created to become a kind of navigator for chemists and help them discover and create materials more economically than with classical approaches. Such savings are obtained by automating most of the initial stages of material synthesis and the ability to synthesize remotely through the cloud, IBM explained.

RoboRXN offers chemical solutions that use machine learning to identify and rank effective and resistant enzymatic reactions. This ultimately provides a more environmentally friendly chemical synthesis. For example, chemists can use this AI system to process vast amounts of data on potentially known enzymes and replace traditional chemical catalysts and toxic solvents with natural compounds derived from plants and vegetables.

The application of solutions in the field of sustainable chemistry is limited by the fact that the substantive knowledge necessary to adapt existing enzymes to new chemical reactions is unstructured. This complicates the process and takes a long time to discover new possible enzymes using traditional experimental methods.

The discovery of enzymes for organic synthesis can provide more efficient, cost-effective and sustainable methods of producing more environmentally friendly products. Enzymes are used in various industries - from the food industry and beverage production to pharmaceuticals and cosmetics. For example, in papermaking, cellulose can be treated with a natural xylanazoic enzyme instead of bleach, which is expensive and pollutes the environment.

• RXN AI for Chemistry technology, the main engine of the RoboRXN system, works on the basis of the conversion method, using neural machine learning to predict the most likely result of a chemical reaction. This is achieved by translating from one "language" (reagents and reagents) to another (products) using character sequences called the Simplified Molecule Representation System (SMILES) to describe chemicals. The optimized synthesis recipes and regulations are then used as a starting point data for the RoboRXN, automated platform for molecule synthesis. The AI system is also equipped with an architecture for retrosynthesis, where instead of predicting the outcome of a possible chemical reaction (direct predictions), the process proceeds in reverse order: the chemicals necessary to create the target molecule are determined.

According to IBM, the accuracy of the RoboRX laboratory in predicting chemical reactions reaches 90% and higher. As of the beginning of December, the platform has more than 29 thousand users and has accumulated more than 5 million predictions for reactions.

2020: Announcement of IBM RoboRXN - AI-platforms for drug creation

At the end of August 2020, IBM introduced a free AI platform for drug creation. A solution called IBM RoboRXN that provides modeling of possible chemical reactions, allows scientists to evaluate the likely benefit of a new drug, which means it accelerates them development and market launch. According to the manufacturer, for identifying and bringing to market a new drug on average requires 10 years and at least $10 million. The goal of IBM RoboRXN is to reduce these costs to one year and $1 million.

IBM RoboRXN combines cloud technology, artificial intelligence, and automation of evaluation of complex reactions of organic chemistry, which can pave the way for the discovery of new drugs. Any chemist connected from home or from work to platform, can offer for evaluation any molecule. RoboRXN will recommend optimal scientific routes and the best source material available on the market. Then RoboRXN self-programmed to "remotely perform experiments in autonomous laboratory. " RoboRXN can use statements from published literature or independently recommend conditions for conducting the experiment.

IBM has released a free AI platform for drug creation

For this task, three models of artificial intelligence were trained. The first model uses retrosynthetic analysis and determines "ingredients," including commercially available precursors. The second and third models focus on synthesis capabilities using a set of data already published in the literature and patents. After completion chemical synthesis process is automatically created analytical a report that will help chemists continue their research.

During the tests, IBM estimated the level of accuracy of the AI platform by 90%, and although black box mystery - no one knows how the AI algorithm takes solutions - still remaining, team trying to increase transparency algorithm to achieve end-user trust.^[1]

Notes