AI in Drug Discovery
Following closely behind the widespread adoption of artificial intelligence (AI) in general, applications of AI in drug discovery have recently reached – and some would say passed – the top of the Gartner hype cycle. Evidence of the market’s excitement can be seen in the recent emergence of a plethora of drug discovery companies citing AI as the main pillar of their business. Simon Smith from BenchSci is maintaining a running list of companies that seem to be crowding this area, currently numbered at 93.
The situation is complicated by the fact that AI is a general term, and is often used loosely to capitalize on its popularity. Because of the general nature of AI, it can be used in many different ways, and for many different purposes, even within the drug discovery world. Of the 93 companies listed by Simon, there are those who have innovated in more conventional spaces, for example, using deep learning (DL) as an efficient tool for virtual screening, not unlike the open source tool presented by Ragoza et al. in 2017. Others have applied machine learning (ML) to enhance traditional QSAR models, or convolutional neural nets specialized to automate image interpretation in high content screening. In fact, as is shown clearly in the list, the overlap between these 93 companies is minimal with many of them solving very different problems in different sectors of drug development.
Being lumped together under the label “AI in Drug Discovery” actually creates a false sense of competition, both for funding (“We need more AI companies in our portfolio”) and for business development (“Management says we need to make a deal with an AI company”). Ironically, this artificial sense of competition may well negate many of the real and perceived benefits obtained from jumping on the AI bandwagon.
At Cyclica, we don’t want to solve just one particular problem really well, with AI or otherwise, although that is how we started out as a company. In our early days, we developed, validated, and commercialized a novel in silico Proteome Screening technology forming the core of our Ligand Express platform. Proteome Screening is a unique computational technique to profile the polypharmacology of drugs and provide broad insights into their mechanism of action, on- or off-target. The method originally was based on biophysical computation alone, although it has since been augmented with ML based methods in the annotation and filtering of results to get to answers more efficiently.
Building on this strong foundation of Proteome Screening, Ligand Express is now being extended both upstream and downstream, enabling researchers to computationally generate new chemical entities (upstream) and investigate the impact of genetic differences on their activity (downstream). Our winning aspiration is to drive drug discovery by empowering scientists in pharma with an integrated cloud-based and AI-augmented platform that enhances how they design, screen, and personalize medicines. We are achieving this vision with Ligand Express, which, we are confident, will eventually be used throughout the pharmaceutical industry to speed up the process using biophysical and AI-augmented computational methods. By putting relevant information at researcher’s fingertips, whether from public sources, in-house proprietary data, or biophysically computed, Ligand Express will enable them to make connections that would otherwise stay hidden, and streamline and even eliminate many of the time-consuming and expensive laboratory experiments currently used for drug discovery.
The following image shows how we will transform Pharma R&D, together with scientists in pharma. It shows a bird’s eye-view of our Map of Drug Discovery Science. At the corners are three important cornerstones of pharmacological research, Genes, Drugs, and Health. The inside of the triangle stands for mechanism, or Understanding, while the outside represents information, or Knowledge. Currently, Ligand Express operates mostly on the inside, as Proteome Screening is based on simulating biophysical mechanism, and Network Analysis is based on mechanistic insights from systems biology. Increasingly, we are improving the predictive accuracy of Ligand Express by augmenting mechanistic computation with AI/ML algorithms that draw on large public and proprietary sets of chemogenomic and systems biology data.
It is well known that genetic data holds remarkable insights into human health; however, scientists are still developing methods to best utilize this plethora of data. With the support of the Ontario Centres of Excellence and IBM, we will soon be introducing a new feature that will drive actionable insights out of genetic data by integrating them within Ligand Express through a technology called Structural Pharmacogenomics (SPGx). Using super computing, Cyclica has prepared a robust genetic database of protein-changing single nucleotide polymorphisms (SNPs) and their locations within protein structure. This mapping allows researchers to quickly identify potential genetic variants that could impact small molecule binding or are implicated in disease.
While the above is certainly important and valuable, a drug needs to do more than simply bind to drug targets; it also needs to have desirable pharmacological properties, particularly regarding absorption, distribution, metabolism, excretion, and toxicity (ADMET). To drive insights in this area, we have built a collection of predictive ADMET models built using proprietary AI methodology to help translate small molecules into drugs. The novel, and patented technology underlying these models has demonstrated enhanced predictive accuracy for generating ADMET predictions compared to traditional classifiers as seen in the validation note here. Understanding polypharmacology through Ligand Express™ and coupling that with ADMET Prediction will allow scientists to make efficient and informed decisions like never before.
ADMET Prediction will be added as an interactive feature to Ligand Express later this year, and is currently available for collaborative projects with partners. Of particular interest for us, and where we believe our partners will benefit the most, is to collaborate on leveraging their extensive proprietary experimental ADMET data with our prediction technology to drive best-in-class pharmacokinetic insights.
Our next step in achieving our vision is an ongoing effort to bring to market a first-in-class Differential Drug Design (DDD) technology, which is currently being prototyped through collaborations with multiple pharma partners. While computer-aided drug design has had many successes, it has been eclipsed for a long time by high throughput screening in combination with large combinatorial libraries. However, the idea of de novo design of molecules for a particular purpose has never really gone away, and DDD is our entry into the field. At its core, it uses an in silico version of the Design, Make, Test, Analyse (DMTA) cycle to find new molecules that will bind to given targets, and not bind to given anti-targets. The result is a drastic reduction in actual in vitro DMTA cycles, with substantial savings in time and resources.
We are excited about the future of the Computational/AI Drug Discovery space, and in particular about our plans to reshape how scientists in the industry use and trust new technologies. We are firm believers in driving user-centric innovation through collaboration, while maintaining a high degree of scientific integrity. This will ensure that we build products that are intuitive, elegant, and insightful, and where the data generated is of high fidelity. By achieving our vision, we will transform pharma R&D, and drive better health outcomes for patients, globally.
Naheed Kurji, President and CEO
Andreas Windemuth, Chief Science Officer
With thanks to our awesome team!