I’ve looked before at the growing role AI is playing in the development of new medicines, whether it’s understanding which compounds to test, or even in the creation of virtual models to test drugs in.

At the forefront of this trend is Insilico Medicine, who you may remember I wrote about recently after they’d developed a system that can guess your age accurately just by looking at you.

They have certainly been busy, and recently published a paper looking at the role of deep learning in predicting the impact drugs might have on the body.  The study saw a neural network trained up to predict the therapeutic use of a huge array of drugs.

Measuring the signals

The team measured the differential signaling pathway activation score for a wide range of different pathways to reduce the deminsionality of the data, whilst ensuring that it remained scientifically relevant.  These were then used to train the neural network.

“The world of artificial intelligence is rapidly evolving and affecting every aspect of our daily life. And soon this progress will be felt in the pharmaceutical industry. We set up the Pharma.AI division to help pharmaceutical companies significantly accelerate their R&D and increase the number of approved drugs, but in the process we came up with over 800 strong hypotheses in oncology, cardiovascular, metabolic and CNS space and started basic validation. We are cautious about making strong statements, but if this approach works, it will uberize the pharmaceutical industry and generate unprecedented number of QALY”, they say.

Pleasingly, the methods developed by the team are open for others to use, and they’ve published their workings out in the paper for others to build upon.  They are currently using the method to develop multimodal neural networks that can predict the properties of a wide range of drugs, molecules and compounds.

“The field of machine learning have recently witnessed an impressive breakthrough in the area of pattern recognition and computer vision. Deep learning, technology to thank for this, continues to disrupt traditional approaches in many other subfields of machine learning. Originally in the 60s, inspired by how the brain works (at least how we understood it back then) deep learning has now developed into a mature engineering concept. The brain however, does not cease to puzzle researchers and, I am sure, contains more sources of inspiration for the future powerful methodologies.”, they continue.

With drugs typically taking over ten years to reach the market and costing over a billion dollars each on average, it seems inevitable that AI will be used to try and bring those costs down.  It’s still too early to say whether this kind of automated approach will become mainstream, but it’s a fascinating area to follow.