Neovulga – Vulgarized Knowledge – Evolutionary algorithms for molecule generation
Neovulga – Vulgarized Knowledge – Evolutionary algorithms for molecule generation
At Neovision, scientific monitoring is key to stay state of the art. Every month, the latest advances are presented to the entire team, whether it is new data sets, a new scientific paper… We screen almost all the news. In our ambition to make AI accessible to everyone, we offer you, every month, a simplified analysis of a technical topic presented by our R&D team.
Today, we will take a look at a scientific paper untitled EvoMol : a flexible and interpretable evolutionary algorithm for unbiased de novo molecular generation, written by Julies Leguy, Thomas Cuachy, Marta Glavatskikh, Béatrice Duval, and Benoit Da Mota.
Background
Molecular generation is a challenge for scientists. From one field to another, such as pharmaceuticals or organic materials, it is difficult to find the same chemical space (i.e. the set of possible molecules). That is why the issues and interests will not be defined in the same way depending on the field.
For example, for organic matter, there is no reliable and representative training data set. The challenge is therefore to find a data-independent algorithm with explicit search spaces, capable of responding to this problem despite this lack.
The authors have therefore sought a flexible solution for the generation of generic molecules. To this end, they designed EvoMol, a molecular generator using an evolutionary algorithm to explore known and unknown areas of a given chemical space using molecular graphs.
However, establishing a diagnosis based on automated recognition implies taking into account several modalities, such as facial expressions, gestures, acoustic characteristics or verbal content. Indeed, an isolated modality rarely provides complete information and each one has its added value.
Presented breakthrough
To understand how EvoMol works, you need to know the logic behind evolutionary algorithms.
These are algorithms whose computational methods are bio-inspired, i.e. based on the observation of nature, and in their case of the theory of evolution. The objective will be, for a given problem, to “evolve” solutions to find the best ones based on mutation and selection mechanisms. Therefore, EvoMol will only generate valid molecular graphs and unique solutions with high scores.
In this paper, the authors explain that they defined a set of seven local and chemical mutations, allowing them to have a large number of possibilities thanks to an extended perimeter of chemical space. This allowed them to test several different targets. In order to follow the exploration process, the authors created a visualization tree that allowed them to easily observe the results.
They were able to generate optimized sets of molecules based on a defined objective. EvoMol achieves excellent performance, especially for classical drug design and molecular materials problems. By working in a controlled way, analyzing mutation by mutation, the method really makes sense from a chemical and business point of view.
Why is it awesome ?
Arthur’s editorial
“Evolutionary algorithms are an AI method inspired by natural evolution. They are used to solve complex search problems with excessive combinatorics. In this paper, the authors use evolutionary algorithms to design chemical compounds”
An evolutionary algorithm like EvolMol is interesting in its ability to work without the need for specific data. This method, which is not well known to the general public, is flexible, allowing it to be easily adapted to different types of problems. Contrary to other AI techniques, here the visualization and interpretation of the results are easy thanks to a visualization tree.
In the long term, and in addition to the advantages presented above, it will be possible to accelerate and facilitate the search for molecules, for example for the creation of new drugs. Moreover, the proposals are based on unique and valid results, so it is easier to select a candidate molecule.
