According to a report published last year, the cost of finding a new medicine is a whopping $2.5 billion, which is up from $1.8 billion in 2010. A popular drug discovery methodology is to use virtual screening where computer programs are used to find a suitable compound that reacts in a way that is desired as chemicals have therapeutic properties. Combing through a large database of chemicals is a daunting task. There is a challenge of massive dimensionality when it comes to analysing structures in drug design, and this is where machine learning comes into the picture.

Machine learning for drug discovery typically involves a training set with known active and known inactive compounds and then finding the probability that a compound is active and then ranking each compound based on its probability of being active.

In order to explore how machine learning can help accelerate the drug discovery process, Google teamed with pharma partners and demonstrated a novel virtual screening method that uses Graph Convolutional Networks(GCNN).

Innovating existing “virtual screening” methods to find potential molecules computationally rather than in a lab is an active area of research. However, the challenges here is to build a method that works well enough across a wide range of chemical space to be useful for finding small molecules with physically verified useful interaction with a protein of interest, i.e., “hits”.

Overview Of The Model

In a recently published work, titled, “Machine learning on DNA-encoded libraries: A new paradigm for hit-finding”, Google’s Accelerated Science team collaborated with X-Chem Pharmaceuticals to demonstrate an effective new method for finding biologically active molecules using a combination of physical screening with DNA-encoded small-molecule libraries and virtual screening using a graph convolutional neural network (GCNN).

Source: https://analyticsindiamag.com/drug-healthcare-deep-learning/