Drug discovery and development is a lengthy, laborious and extremely expensive process. In recent years, many biological databases have been established to curate the structural and functional properties of bioactive molecules published in literature. With artificial intelligence (AI) making huge breakthroughs in a wide range of applications, pharmaceutical companies are now using cutting-edge AI, capable of learning from these biological databases, to enhance and accelerate their drug development process.


Deep learning is a branch of machine learning that uses deeply connected artificial neural networks (ANNs) to perform complex tasks. They perform exceptionally well at extracting key features from complex datasets. Deep learning has made significant breakthroughs in numerous applications, including computer vision, speech recognition, natural language processing and even in the ancient Chinese board game Go.

ANNs could be used to extract important structural, functional and biochemical features from existing drugs. There are many types of biological databases, including genomic, transcriptomic, proteomic, metabonomic, structural, drug-target interaction, clinical trials, drug responses, drug efficacy, etc. The advantage of using deep learning for drug development is that ANNs can utilize several biological databases to extract key features of drugs, and in turn generate new drugs using those features. Instead of screening an astronomical number of molecules for a viable drug candidate, ANNs can be used to just generate one.

Generative Adversarial Networks (GANs) are a class of ANNs that can learn to generate new drugs. GANs are made up of two main components; a generator neural network and a discriminator neural network. Using Gaussian noise as input, the generator randomly generates a new drug, defined by a combination of structural, functional and biochemical features. The discriminator then compares this new drug to an existing drug and attempts to distinguish the real one from the fake one. After many iterations, the discriminator becomes increasingly better at discriminating between real drugs and fake drugs. However, this causes the generator to also become more sophisticated, generating new drugs that are very difficult to distinguish from existing drugs.

Through this iterative back and forth learning process, both the generator and discriminator will learn key structural, functional and biochemical features. Once the GANs has finished training, the generator neural network can create drugs de novo by sampling the Gaussian distribution as input. These drug candidates can then be screened for synthesis, followed by experimental testing, and eventually clinical trials, going through the entire drug development process.

There is an increasing amount of investment from pharmaceutical companies using AI for drug development. In 2017, Sanofi signed a €250 million deal with an AI based small molecule drug discovery startup Exscientia. GlaxoSmithKline (GSK) also signed a similar deal with Exscientia for $43 million. Earlier this March, Atomwise raised a $45 million Series A. Later in April, BenevolentAI raised $115 million from investors, which increased their valuation to $2 billion. It is an exciting time for deep learning in pharmaceutical research, which has led to huge investments into AI startups for drug discovery and development.

Developing a novel drug is complicated and expensive. However, state of the art AI can be used to tackle major challenges to enhance and accelerate the drug development process. Further advancements in deep learning research will ultimately lead to development of more effective drugs and a better understanding of health and disease.


References:

  1. Goodfellow, Ian J., Pouget-Abadie, Jean, Mirza, Mehdi, Xu, Bing, Warde-Farley, David, Ozair, Sherjil, Courville, Aaron C., and Bengio, Yoshua. Generative adversarial nets. NIPS, 2014. 
  2. Nikolenko, S. Creating Molecules from Scratch I: Drug Discovery with Generative Adversarial Networks. https://medium.com/neuromation-io-blog/creating-molecules-from-scratch-i-drug-discovery-with-generative-adversarial-networks-9d42cc496fc6. Neuromation (2018).
  3. Shu, C. Atomwise, which uses AI to improve drug discovery, raises $45M Series A. https://techcrunch.com/2018/03/07/atomwise-which-uses-ai-to-improve-drug-discovery-raises-45m-series-a/Tech Crunch (2018).
  4. BenevolentAI raises $115 million to extend its leading global position in the field of AI enabled drug development. https://benevolent.ai/news/announcements/benevolentai-raises-115m-for-ai-enabled-drug-development/ . BenevolentAI (2018).
  5. GSK Launches Up-to-$43M AI-Focused Collaboration with Exscientia. https://www.genengnews.com/gen-news-highlights/gsk-launches-up-to-43m-ai-focused-collaboration-with-exscientia/81254606Genetic Engineering & Biotechnology News (2017).
  6. Sanofi, Exscientia Ink Up to €250M Deal for Bispecific Drugs Against Metabolic Diseases. https://www.genengnews.com/gen-news-highlights/sanofi-exscientia-ink-up-to-250m-deal-for-bispecific-drugs-against-metabolic-diseases/81254318. Genetic Engineering & Biotechnology News (2017).
The following two tabs change content below.

Anthony Zhao

Previous post The Telomere Clock: Theory of Aging
Next post Does Adderall Pass the Test?

Leave a Reply

Your email address will not be published. Required fields are marked *

Close

Feed currently unavailable. Check us out on Twitter @immpressmag for more.


Sponsors