How AI Can Speed Up Drug Discovery

The US FDA defines five steps for the development of a new drug: discovery and development, preclinical research, clinical research, FDA review, and FDA post-market safety monitoring.


The US FDA defines five steps for the development of a new drug: discovery and development, preclinical research, clinical research, FDA review, and FDA post-market safety monitoring.[1] The Tufts Center for the Study of Drug Development estimates the average cost of developing a new drug at US$2.55 billion, with the process potentially taking more than 10 years.[2]

The first step, drug discovery, typically involves one of four scenarios: finding new insights into a disease, finding possible effects of a drug by testing molecular compounds, repurposing existing drugs, or manipulating genetic materials. At the drug discovery stage, thousands of compounds may be potential candidates for development as a medication.[3] They need to go through a series of tests and only a small number will advance to further research.

To accelerate drug discovery and reduce the costs of drug development, pharmaceutical companies are introducing AI technologies such as machine learning and deep learning into the processes.

How AI Can Speed Up Drug Discovery

Finding new insights into a disease

Drug discovery is a data-driven environment with a massive of data such as high-resolution medical images, genomic profiles, metabolites, molecular structures, and biological information.[4] This information is published in papers and journals, however it can be a challenge for researchers to keep up with it. AI can use machine learning and deep learning to correlate, assimilate, and connect existing data more rapidly in order to help discover patterns in the data pools. By reviewing scientific research papers, AI can make connections that provide possible hypotheses for drug discovery.

  • Considering the huge volume of published medical papers, and that human researchers typically read only between 200 to 300 articles per year, having machines do the reading can bring higher efficiency to the process. Pharmaceutical giant Pfizer uses the IBM Watson for Drug Discovery cloud-based platform, which comprises 25 million Medline article abstracts and one million medical journal articles, in its immuno-oncology research.[5]
  • British company BenevolentAI is using AI and deep learning to mine and analyze vast quantities of complex scientific information to accelerate drug discovery. Scientific papers, clinical trials information, and datasets can be digested to provide new insights for researchers. With this approach, BenevolentAI can shorten the process and predict how the compounds can be more efficient targeting diseases.[6] For instance, BenevolentAI used AI technology to help identify a potential hypotheses for the treatment of ALS (also known as Motor Neurone Disease) in 2016.

Finding possible effects of a drug by testing molecular compounds

Finding new compounds for a medicine is difficult because the possible combinations are countless. Such research requires medical data on genes, proteins, metabolites, molecular structures, and biological information.[7] Processing this huge amount of information can be a very time-consuming task. Pharmaceutical companies are discovering that AI techniques such as deep learning algorithms can process the same information much faster.

  • Exscientia’s AI platform encodes deep-rooted knowledge for compound design and assessment, to screen compounds in cells or animal models. By comparing the results of a newly designed compound with the anticipated performance and with other molecules, researchers are able to evolve compound designs to help drug discovery. Exscientia can rapidly synthesize and assay small batches of compounds, which could help refine the models being developed and evolve the designs.[8]
  • Atomwise’s drug research technology, AtomNet, uses deep learning algorithms and elastic supercomputer platforms to screen more than ten million compounds daily, which speeds up the process of identifying molecules as potential drug candidates. AtomNet’s molecule simulations are more effective than physical high-throughput screening methods, and can also provide a better understanding of the toxicity, side effects, and efficacy of a drug.[9]
image (31).png
Molecular structure [10]

Repurposing existing drugs

Drug repurposing, also known as drug repositioning or therapeutic switching, is the application of known drugs and compounds to treat new indications.[11] One advantage of drug repositioning is that most of the repositioned drugs have already passed a series of tests, and so have less risk of unexpected toxicity or side effects. With the help of machine learning algorithms, pharmaceutical companies can repurpose drugs faster and at lower costs than developing new drugs.

  • IBM collaborates with Teva Pharmaceuticals in drug repurposing efforts by using real-world data and machine learning algorithms. IBM Watson uses cognitive technologies to mine unstructured data and explore the relationships between drug molecules and particular diseases, which has the potential to significantly improve drug repurposing.[12]
  • Bioinformatics company NuMedii works with Astellas Pharma on drug repurposing projects using machine learning techniques. NuMedii uses neural network-based algorithms to find novel drug candidates from its big data resources, using biological, pharmacological and clinical data points. NuMedii can thus help repurpose existing drugs or create drug candidates for other medical indications.[13]

Manipulating genetic materials

Manipulating genetic materials for drug discovery is also known as personalized medicine or precision medicine. This kind of drug discovery can be more effective in treatment because it is based on individual health data paired with predictive analytics.[14] In order to efficiently gather, analyze, store, and trace a person’s detailed information, especially when the data is huge and unstructured, pharmaceutical companies use deep learning, machine learning, or computer vision.

  • MIT Clinical Machine Learning Group’s precision medicine research is focused on unsupervised learning, deep learning, time-series modeling, approximate probabilistic inference, structured prediction, and semi-supervised learning algorithms for natural language processing.[15] The research group uses these technologies to better understand disease processes and design drugs for the treatment of diseases such as Type 2 Diabetes.
  • An AI-driven biotechnology company, Recursion Pharmaceuticals works with global biopharmaceutical company Sanofi to identify uses for clinical stage molecules in the treatment of genetic diseases. Recursion uses computer vision to do image analysis of individual cells, screens them across a library of genetic disease models, and uses machine learning technology to derive new indications.[13]
image (32).png
DNA Strand [16]

The Future of Drug Discovery

We know that artificial intelligence can be applied in drug discovery to make the process faster. There are also areas where artificial intelligence can help in drug development. Clinical trials, for example, are currently classified into five phases and usually require more than three thousand test subjects to proceed from phase one to phase three.[17] Most pharmaceutical companies use recruitment firms to find clinical trial subjects by examining individual medical records.[18] This task takes time and the efficacy is low. Companies can use machine learning to train a model that includes age, sex, treatment history, and current health status to build an inclusion/exclusion criteria that will speed up this aspect of clinical trials research.

Moreover, AI can help test the side effects or toxicities of candidate drugs. Cyclica, a Canadian startup, uses a suite of computational algorithms to evaluate and predict how drugs might interact with the human body.[19] This kind of testing helps pharmaceutical companies identify a drug candidate’s side effects before clinical trials, so companies can make corrective adjustments in advance.

Drug discovery can benefit from machine learning, deep learning, and computer modeling. However, there is also the potential for introduction of biases from unbalanced data, which might cause errors or discrimination while AI is training the neural networks.[8] A research team from Insilico Medicine discovered that accuracy could become unstable unless the neural network had been trained using diverse datasets. The range, quantity and quality of input data is therefore a key factor for further implementation of AI in drug discovery.


[1] The Drug Development Process:

[2] Tufts CSDD R&D Cost Study Now Published:

[3] Discovery and Development:

[4] AI Provides New Insights for Accelerated Drug Development:

[5] AI in Pharma and Biomedicine – Analysis of the Top 5 Global Drug Companies:

[6] What if AI could take your research to the next level?:

[7] Machine Learning Drug Discovery Applications – Pfizer, Roche, GSK, and More:

[8] Artificial Intelligence: will it change the way drugs are discovered?:

[9] AI start-up Atomwise raises $45m to fund drug research technology:

[10] Picture of a molecular structure:

[11] Drug repositioning:

[12] IBM, Teva to Use A.I. for Drug Repurposing Program:

[13] How Pharmaceutical And Biotech Companies Go About Applying Artificial Intelligence in R&D:

[14] Personalized medicine:

[15] MIT research:

[16] Picture of Genetics:

[17] Clinical trial:


[19] Artificial Intelligence Meets Drug Development:

Analyst: Paul Fan| Editor: Michael Sarazen

0 comments on “How AI Can Speed Up Drug Discovery

Leave a Reply

Your email address will not be published.

%d bloggers like this: