AI for drug discovery: Our focus on emerging infectious diseases

How drugs are typically discovered

The discovery of new medicines is a key pillar underpinning modern medicine. The impacts of critical life threatening and life limiting diseases have been significantly reduced through humanity’s ability to discover safe and effective drugs.

The conventional drug discovery process involves the following key steps:

• Target identification – identification of new targets associated with a disease such as proteins that can be inhibited or upregulated.
• Hit identification – experimental and computational screening of hundreds of thousands to millions of compounds using high throughput assays to test activity against the target in vitro. Cell-based assays also assess the toxicity of these hit compounds.
• Hit-to-lead development – systematic synthetic alterations are made to the hit compounds to improve potency and confer more “drug like” properties by establishing Structure Activity Relationships (SAR). This process typically involves hundreds of compounds. Lead compounds with greater activity and selectivity, and reduced toxicity, are prioritised for optimisation.
• Lead optimisation and candidate selection – Compounds are further synthetically modified and progressed to preclinical assessment, usually in both healthy and diseased animal experiments. Compounds with the best pharmacological profiles, safety, and effectiveness are prioritised. Less than 10 compounds are selected as suitable for entering human clinical trials.
• Phase I clinical trials – the drug candidates are administered to 20-80 healthy human volunteers. The subjects are monitored closely to look for any toxicity issues or safety concerns.
• Phase II clinical trials – drug candidates that pass Phase I trials are administered to several hundred volunteers who have the disease in question. This trial is primarily to assess the effectiveness of the drug in treating the disease, but safety profiles and side effects are also monitored.
• Phase III clinical trials – drug candidates are administered to thousands of volunteer patients to demonstrate safety and compare efficacy against existing treatments on a larger population, typically 300 to 3,000 volunteers.
• Approval and Phase IV monitoring – If successful, a drug candidate will be approved as a drug by the relevant national regulatory body. Phase IV involves continuous monitoring of any safety issues and effectiveness once the drug is administered through the healthcare system.

This is a lengthy, costly process that limits our ability to respond to emerging health threats. We believe AI can be the key to accelerating the discovery of new medicines, both in early discovery efforts and in selecting candidates more likely to succeed through clinical trials.

Our focus on antimicrobial resistance

Imagine a world where common infections can no longer be treated by existing drugs in our hospitals and clinics. Prior to the discovery of antibiotics, our life expectancy was around 40 years, and common occurrences such as surgery, childbirth, and traumatic injury were heavily complicated by infection.

Today, key human pathogens such as MRSA and E. coli are becoming increasingly resistant to antibiotics. This is known as antimicrobial resistance and is reducing the effectiveness of all known antibiotic classes. Antimicrobial resistant infections killed 1.27 million people in 2019 and are on track to kill 10 million people annually by 2050, overtaking the number of deaths from all cancers. At the same time, fewer and fewer new antibiotics are being discovered and entering the market.

The predicted global impact of antimicrobial resistance by 2050:

• Deaths from infections that were previously treatable with antibiotics will exceed 10 million per year
• AMR will result in up to 7.5% global decrease in livestock production
• Global GDP will decline by 3.8-5%
• The number of people in extreme poverty will increase by 28.3 million
• Global real exports will shrink by 1.1%
• Global healthcare costs will increase from $300 billion to >$1 trillion per year.

Consequently, we urgently need new and more rapid approaches to discovering antibiotics. The machine learning tools we are developing can help remove inherent biases and allow the investigation of much broader chemical space to enable discovery of completely new antibiotic classes to circumvent existing and emerging resistance.

Our focus on pandemic preparedness

Viral pathogens with pandemic potential are a significant threat to our healthcare systems and economy. For example, SARS-CoV-2, the virus responsible for COVID-19, already accounted for 6.5 million deaths worldwide at the time of writing and has had devastating impacts on global economies and supply chains. Alongside vaccines, antivirals are a key technology for limiting the impact of a future pandemic. 

Focusing on zoonotic diseases, which transmit from animals to humans:

• 75% of emerging human diseases are zoonotic (animal to human)
• 300% increase in outbreaks seen in zoonotic diseases over the past 30 years
• ~2.5 billion cases of illness and 2.7 million deaths from zoonoses every year
• 540,000 – 850,000 unknown viruses existing in nature are estimated to be zoonotic

Unlike AMR, we are unaware of the identity of a future pandemic-causing zoonotic disease. However, our ability to rapidly develop antivirals in such an event will be aided by the tools we are developing today. The ability to rapidly screen a larger pool of compounds against new targets, aided by new AI methods such that successful candidates can be readily deployed at the onset of a new pandemic, could potentially reduce the severity of infection by such a pathogen.

We envision a future where data science is deeply imbedded into the drug discovery process, giving us rapid and cost-effective access to AI-derived next generation medicines for a range of currently untreatable diseases.