One of the shared and fundamental objectives of most researchers in chemistry is the need to predict the properties of a molecule, such as its boiling or fusion point. Once the researchers can determine this prediction, they are able to move forward with their work which gives discoveries that lead to drugs, materials, etc. Historically, however, traditional methods of unveiling these predictions are associated with a significant cost – by spending time and wear on equipment, in addition to funds.
Enter a branch of artificial intelligence known as automatic learning (ML). ML has reduced the load of predicting the properties of molecules to a certain extent, but the advanced tools which most effectively accelerate the process – by learning existing data to make rapid predictions for new molecules – require that the user has a significant level of programming expertise. This creates an accessibility barrier for many chemists, who may not have significant IT competence required to navigate the prediction pipeline.
To alleviate this challenge, researchers McGuire research group MIT created ChimxploremlA friendly office application that helps chemists make these critical predictions without requiring advanced programming skills. Available free of charge, easy to download and functional on traditional platforms, this application is also designed to work entirely offline, which helps keep the search data owners. The new exciting technology is described in a Article published recently in the Journal of Chemical Information and Modeling.
A specific obstacle in learning the chemical machine consists in translating molecular structures into a digital language that computers can understand. Chemxploreml automates this complex process with powerful and integrated “molecular interests” that transform chemical structures into informative digital vectors. Then, the software implements advanced algorithms to identify the models and predict molecular properties with precision such as boiling and fusion points, throughout an intuitive interactive graphical interface.
“Chemxploreml's objective is to democratize the use of automatic learning in the chemical sciences”, explains Aravindh Nivas Marimuthuhu, postdoc of the McGuire group and the main author of the article. “By creating an intuitive, powerful and offline office application, we directly put predictive modeling in the hands of chemists, regardless of their programming context. This work not only accelerates the search for new drugs and materials by making the screening process more quickly, but its flexible design also opens doors for future innovations. ”
Chemxploreml is designed to evolve over time, so that future techniques and algorithms are developed, they can be integrated transparently into the application, ensuring that researchers are always able to access and implement the most recent methods. The application was tested on five key molecular properties of organic compounds – melting point, boiling point, steam pressure, critical temperature and critical pressure – and reached high precision scores up to 93% for critical temperature. The researchers also demonstrated that a new more compact method to represent the molecules (Vibration) was almost as precise as standard methods, such as Mol2vec, but was up to 10 times faster.
“We are considering a future where any researcher can easily personalize and apply automatic learning to solve unique challenges, the development of sustainable materials to exploration of complex chemistry of interstellar space,” explains Marimuthhu. The senior author and the deputy professor of career development chemistry of 1943, Brett McGuire.