AI tissue analysis enables rapid discovery of drug candidates
AI could address the most significant issue in drug discovery, finds Ben Hargreaves, and also offers the potential to use tissue samples to create precision medicine for specific patient groups.
The scale of the challenge
Cancer is the second leading cause of death worldwide, accounting for nearly 10 million deaths in 2020, according to the World Health Organization. In 2020, the US National Cancer institute had a budget of $6.4 billion for research into the disease, which represents a fraction of the overall spend dedicated to learning more about cancer’s development in humans and in the creation of new treatments. Research published this year suggests that the incidence of Europeans dying from cancer is likely to increase by 32% in 2040, compared to 2020. This is being driven by an ageing population but will only increase the pressure on society to find better treatments and to better understand the progress of the disease.
One of the positives in the fight against cancer is that even though cases are likely to rise over time, the capability of technology will also continue to increase. An area that offers hope to accelerate an understanding of cancer and speed up drug development is artificial intelligence. The potential of the area has motivated an increasing number of companies to emerge in the space, with a greater number of collaborations occurring between these AI-specialists and big pharma companies.
Owkin is a French-American startup that is using AI to discover and develop treatments, with a particular focus on cancer. The team recently partnered with the Francis Crick Institute and The Royal Marsden NHS Foundation Trust to conduct research into the evolution of kidney cancer tumours and how there are difference in its microscopic structure. The aim is to help doctors provide more effective treatments to patients, as cases of kidney cancer continue to increase due to an ageing population, rising rates of obesity, and the impact of smoking.
AI could provide a valuable tool in this instance as treatments that fail to be effective could be influenced by intratumour heterogeneity, where distinct tumour cell populations within a tumour possess different molecular and phenotypical profiles.
A spokesperson for Owkin explained how their work will be carried out, in practice: “We are using AI to predict tumour evolution based on histology slides (over 1,000 tissue samples from 100 different tumours). By finding a way to predict unique evolutionary features in every patient, we can then predict outcomes, allowing doctors to tailor their treatment.”
If doctors are given the tools to predict a patient’s outcomes then they are better able to tailor treatments to suit individual needs, the spokesperson continued. The use of low-cost AI tools on digital pathology slides, rather than large-scale genomic sequencing, could provide information helpful to the day-to-day management of patients’ treatments. The larger aims of the study, meanwhile, are to “discover invaluable insights into the interindividual differences in tumour evolution, progression, and treatment resistance,” the spokesperson stated.
Potential in drug discovery
Owkin’s tools and business extend beyond being able to provide insights for the future treatment of kidney cancer. Using its AI platform and knowledge of tumour evolution, the startup is able to interpret histogenomic biomarkers to discover and rank genes and proteins with drug target potential.
With drug development having a failure rate of 96% and costing potentially billions in the process, the pharmaceutical industry is desperate to find a way to discover potential targets in a more efficient manner. This is why Owkin has already partnered with big pharma companies, such as Sanofi and Bristol Myers Squibb, and why there are estimates that the industry’s overall spend on AI drug discovery could exceed $3 billion by 2025.
Owkin drug discovery strategy is to utilise ‘reverse translation,’ by starting with patient data to delineate the difference in treatment outcomes between individuals, using multimodal data to tackle the heterogeneity.
From there, “We then build a platform discovery pipeline allowing the use of multimodal models to predict patient clinical outcomes to define more homogeneous patient subgroups and identify histogenomic biomarkers of interest,” the spokesperson explained. From the biomarkers that are identified, the company then selects which drug targets have the most potential.
Exscientia, another AI drug discovery company that has entered a deal with Sanofi and has been working in the space for a decade. Exscientia CEO Andrew Hopkins explained why AI technology represents an advance on traditional methods: “At Exscientia, we have repeatedly demonstrated our ability to create novel, optimised drug candidates several years faster than the industry average. By applying AI to drug design, new drugs coming through our pipeline have taken only 12 to 15 months from starting the design project to identifying a drug candidate.”
By comparison, the traditional approach to drug discovery would take approximately four and a half years. Such a reduction in time is also able to reduce the cost of development, allowing more clinical candidates to pass through into trials.
Similarly to Owkin, Exscientia is able to use live patient tissue in its AI-driven drug discovery process, enabling the company to use the “closest representation of the patient to help inform every stage of drug discovery and development,” the company’s spokesperson said.
In terms of the process, the spokesperson outlined: “Our automated microscopy tools allow us to analyse biological responses in these complex specimens with unprecedented speed and single-cell resolution, and we can do this without fully dissociating a tissue into individual cells, thus maintaining vital parts of the tumour microenvironment and cell viability.”
For Exscientia, its approach has already seen success as its drug candidate, EXS-21546, became the first AI-designed molecule for immuno-oncology to enter human trials, after being discovered in just nine months. To develop this particular candidate, the company designed and eliminated 175 novel candidates. This was made possible through the use of its AI-driven platform, which is able to generate candidates at a faster rate, and then the novel candidates can be checked for key aspects of the target product profile, such as limited adverse effects. EXS-21546 will enter Phase Ib/II studies later this year, while the company is also using its technology to identify patient signatures and potential biomarkers to determine which patients will respond best to the treatment.
Currently, the company is working on building out a 26,000-square-foot robotic laboratory in Oxfordshire, UK, in which it plans to automate the chemistry and biology utilised in the drug discovery process. The eventual aim is to have the capability to create drugs designed by AI and made by robots.
In the long-term, this offers a vision of drug treatment where patient tissues samples could be taken and a precision medicine could be engineered from the analysis of drug effects on the samples, before it is then produced by an automated robotics system. Of course, this is not the reality today but as advancements are made by companies such as Owkin and Exscientia, perhaps it could be. Regardless, in the short-term, the AI work being done to generate a greater number of drug candidates is getting closer and closer to gaining a drug approval and, when this happens, interest in the area is likely to accelerate even further.