Reece Armstrong speaks to Natalia Mateu, Co-Founder, Jelena Aleksic, Co-Founder and Peter Curran, Chemoinformatics Scientist, about utilising 3D chemistry and AI to get small molecule therapies from the clinic to the bedside.
RA: How is Pharmenable aiming to replicate the specificity of biologics in small molecule drugs?
NM: Biologics are target-specific. This specificity is beneficial, as it decreases off-target interactions which are usually involved in the side effects of a medication. But biologics are big molecules, they cannot cross membranes and cell barriers. Consequently, they are unable to target intracellular space. They also have some pharmacological and production disadvantages compared to small molecules, such as a very long half-life and limited administration routes (mostly intravenous). Small molecules are able to access intracellular targets, are orally available and pose overall fewer manufacturing constraints.
Our proprietary drug discovery platform provides a powerful engine for generating novel, safe and effective chemical entities that are target specific. By applying guided de novo molecular design at the very beginning of the hit discovery process, we ensure that specific and powerful small molecule drug candidates are selected from the start. Using our platform, we model targets to identify their key molecular interactions, and use AI to identify and expand the most relevant parts of chemical space for drugging that specific target. Advanced medicinal chemistry expertise guides the development of synthesisable, complex and three-dimensional small molecules, with increased specificity and reduced off-target effects.
RA: You’re approaching small molecule design using synthesisable 3D chemistry? What advantage does this approach have over traditional approaches?
NM: Looking at successful clinical molecules can provide clues on what attributes may improve the ratio of compounds that reach the clinic and eventually benefit patients. Many successful molecules are of natural origin, which tend to have three- dimensional structures and higher complexity. For instance, having a chiral centre is often used as a measure of complexity. Chiral carbons are commonly associated with more effective and selective binding to target proteins.
The presence of chiral carbons opens chemical space, obtaining novel molecules with enhanced tridimensionality, and increases potency and selectivity which are highly sought-after attributes for molecules to be developed into medicines. Compared to 2D small molecules, 3D molecules that are complex typically translate into decreased secondary effects, as well as reduced toxicity, which is of utmost importance when considering patient benefits.
RA: What disease areas are you focused on and how can an AI approach help generate novel drugs for these areas?
NM: We specialise in challenging targets, where access to novel chemical space and complex, 3D molecules provide advantages in comparison to other approaches. Our in-house portfolio of drugs focuses on oncology, finding new solutions for areas of high unmet clinical need. For example, one of the applications where we use our technology and the specificity of our molecules is tackling specific protein- protein interactions (PPIs) of major cancer targets.
In addition, we optimise the properties of our molecules to ensure that they are good drug candidates for specific therapeutic areas and their requirements. For instance, crossing the blood brain barrier is key for neurodegenerative therapy, and it is usually impenetrable by biologic drugs. Some of our partnerships focus on neurodegenerative disease, with many compelling reasons: a strong scientific rationale, an enormous potential for small molecules, a patient population that keeps increasing and is in acute need of therapeutics, and not uncommonly some members of our team also have personal stories that drive them to make a difference in this therapeutic area.
Our drug discovery platform is target-agnostic; and so is the AI contribution to drug discovery in general. AI can remove some of the biases present until now and help with generating endless possibilities and chemical space to explore. This process, however, has to be guided by human scientific expertise.
RA: What are the limitations of AI in drug discovery?
PC: AI/ML methods are being applied in a variety of ways to all stages of the drug discovery process – from understanding diseases through AI-based systems biology approaches, to generating novel molecular entities in hit discovery, and to the selection of patient cohorts in clinical trials and many more.
A common challenge across the field is that models are only as good as the data used to train them. The availability of high-quality, curated data is often a limitation, particularly data from in vivo models and later stages of drug discovery, as is the need for both positive and negative data. This important issue is being tackled through a number of different open data initiatives. In addition, even with the right data being available, it remains critical to ask the right questions and understand the answers – scientific rigour and expert human input are critical in order for the application of AI to yield genuine insights.
Within small molecule hit discovery specifically, AI-enabled de-novo design has the potential to unlock significantly more unexplored chemical space than the more traditional virtual screening approaches, but the synthesisability of the predicted molecules can be a challenge. We tackle this issue using chemUNIVERSE, our proprietary human-curated repository of medicinal chemistry knowledge. We use chemUNIVERSE in combination with AI- based methods to build up dynamically generated collections of billions of molecules that are both target relevant and synthesisable, while maximising chemical diversity. This maintains synthesisability whilst accessing state-of-the-art novel chemistry.
RA: PharmEnable is focused on GPCR targets in neurology. Could you discuss some of the challenges of working in this area?
NM: GPCRs are large, complex proteins and are often challenging to modulate with a small molecule, particularly one that has properties suitable for development as a therapeutic agent for neurological disease. The challenges are exacerbated due to the difficulty finding selective small molecules with balanced CNS drug-like properties.
The partnership between Sosei Heptares and PharmEnable is highly complementary. PharmEnable brings the power of combining AI-enabled drug discovery with human expertise to deliver innovative small molecule drugs against challenging biological targets. Sosei Heptares has extensive knowledge around the GPCR target, as well as technologies and skills that can elucidate the structure of challenging GPCRs. Our approach in tackling this problem uses AI and machine learning with advanced medicinal chemistry expertise to map chemical space for a given biological target, identifying target-relevant regions and populating them with novel three-dimensional and complex molecules. In doing so, we enable the precise design of potential drug candidates with increased selectivity and improved drug-like properties.
Volume 23 – Issue 4, Fall 2022
Natalia Mateu, Co-Founder and CSO: Mateu brings 15 years of organic and medicinal chemistry experience within the pharmaceutical sector and academia, from the University of Cambridge and Janssen/J&J.
Jelena Aleksic, Co-Founder and CBO: A geneticist and serial entrepreneur, Aleksic brings significant commercial and business development experience and is leading PharmEnable’s business and partnership efforts.
Peter Curran, Chemoinformatics Scientist: A cheminformatician by background, at PharmEnable Curran is developing tools and systems to enable the rapid exploration of synthetically accessible chemical space.