Creating a better drug
Problem being addressed
The process of drug discovery is expensive and time-consuming, and its duration is mostly determined by the number of cycles of design, synthesis, and testing needed to improve the properties of successive generations of small molecules to match the various criteria needed in a drug discovery project.
The use of deep learning technologies could augment the typical practice of using human intuition in the design cycle, and thereby expedite drug discovery projects. The proposed model is a deep neural network model that advances the state of the art in machine learning approaches to molecular design. In its typical use, the model network is modified with transfer learning (learning of downstream tasks with additional training on small, specialized datasets), and then proposes a sequence of molecules that are chemically similar to the input but likely improved for specified properties of interest.
Advantages of this solution
The presented model yields a 77% lower failure rate compared to state-of-the-art models and succeeds on a number of downstream drug discovery tasks.
Solution originally applied in these industries
Possible New Application of the Work
Chemical and Materials Industry
Applying a deep neural network model that advances the state of the art in machine learning approaches to molecular design in chemical industry can significantly speed up the process and make the R&D stage cheaper.
Although the main examples of the model application involved making improvements to existing molecules, it could also be used early in the drug-discovery process to speed up the breakthrough in such areas as HIV or cancer treatment.
Source URL: #############