BioAscent’s cutting-edge computational platform, combined with our extensive expertise in medicinal and synthetic chemistry, decodes the complexities of the blood-brain-barrier (BBB), facilitating the prediction of drug permeability. We transform customers’ CNS drug candidates from hopefuls to high-potential contenders, accelerating the success of CNS drug discovery projects.
Central Nervous System (CNS) disorders present a formidable challenge in drug discovery due to their inherent complexity. These disorders often arise from intricate interplay between genetic, environmental, and lifestyle factors, resulting in multifaceted pathologies that affect diverse neural circuits and signaling pathways. The heterogeneity of CNS disorders, even within a single diagnosis, further complicates therapeutic development, as patients may exhibit varied responses to treatment. Moreover, the blood-brain barrier (BBB) poses a substantial hurdle, limiting the access of potential drugs to their target sites within the brain. These challenges necessitate innovative approaches that can navigate the complexities of CNS biology and facilitate the discovery of effective therapeutics.
The BBB (Blood-Brain Barrier) is a highly selective semipermeable border of cells that prevents harmful substances in the blood from entering the brain. It's essential for maintaining a stable environment for the brain to function, but it also presents a significant challenge to small molecule penetration due to several key factors:
Tight Junctions: The endothelial cells lining the blood vessels in the brain are connected by tight junctions, forming a highly selective barrier that limits the passage of molecules between cells. This restricts the movement of most molecules unless they have specific transporters or can pass through the lipid bilayer.
Efflux Transporters: The BBB expresses various efflux transporters, such as P-glycoprotein (P-gp), which actively pump drugs and other molecules out of the brain, further limiting their access.
Limited Passive Diffusion: While small lipophilic molecules can passively diffuse across the BBB, this process is often inefficient. The BBB's lipid bilayer is tightly packed and presents a significant barrier to larger or more polar molecules.
Low Endocytosis: Endocytosis, the process by which cells engulf molecules, is limited at the BBB. This restricts the entry of molecules that rely on this mechanism for transport.
Metabolic Enzymes: The BBB also expresses enzymes that can metabolise drugs, reducing their concentration before they reach the brain.
Physicochemical Properties: A molecule's physicochemical properties, such as size, charge, lipophilicity, and hydrogen bonding potential, significantly influence its ability to cross the BBB. Molecules that are too large, too polar, or have high hydrogen bonding potential are less likely to penetrate the BBB.
Disease States: In certain disease states, such as Alzheimer's disease or brain tumours, the integrity of the BBB can be compromised, leading to increased permeability. However, this can also make it more difficult to predict drug delivery and may result in unwanted side effects.
The BBB's complex structure and active mechanisms create a formidable obstacle for drug delivery to the brain. Successfully designing small molecules that can effectively penetrate the BBB requires careful consideration of these factors and the development of strategies to overcome them. This includes optimising physicochemical properties, utilising transporter-mediated uptake, and developing novel drug delivery systems.
Our goal is to empower customers in their CNS drug discovery activities. We recognise that each project has unique requirements, and we work collaboratively with our customers to tailor our computational approaches to their specific needs. Our team has a proven track record of success in applying these methods to optimise CNS drug candidates. For example, we recently collaborated with a customer to identify a lead compound with significantly improved BBB penetration (Kp,uu 0.35) compared to the precursor compound (0.16) in just one cycle of design, ultimately accelerating their preclinical development timeline.
By partnering with BioAscent, customers gain access to a dedicated team of experts passionate about accelerating CNS drug discovery. We are confident that our comprehensive computational toolkit and customer-focused approach can significantly enhance the success of CNS drug development programmes.
Our in silico team at BioAscent boasts a comprehensive suite of tools and expertise for optimising central nervous system (CNS) drug candidates, with a particular focus on enhancing brain penetration. We employ a multi-pronged approach, leveraging both well-established and cutting-edge techniques to comprehensively assess and predict a molecule's ability to cross the blood-brain barrier (BBB).
CNS MPO Score: This multi-parameter optimisation score utilises a desirability function-based approach, integrating key physicochemical properties such as lipophilicity (cLogP), distribution coefficient (cLogD), molecular weight, topological polar surface area (TPSA), hydrogen bond donors/acceptors, and pKa. By assessing the alignment of these properties with known CNS drugs, the CNS MPO score provides a rapid and intuitive assessment of a compound's overall potential for CNS penetration.
BBB Score: This model employs stepwise and polynomial piecewise functions to predict BBB penetration based on five carefully selected physicochemical descriptors. The model's parameters are trained on experimental data, offering a quantitative prediction of a molecule's likelihood of crossing the BBB.
Free Energy of Solvation (E-sol): The unbound brain-to-plasma partition coefficient (Kp,uu) represents the ratio of unbound (free) drug concentrations in the brain and plasma. It's a crucial parameter in CNS drug discovery, as it directly reflects a drug's ability to cross the blood-brain barrier and reach its target site in the brain. By quantifying the extent to which a drug can partition into the brain, Kp,uu provides a key metric for evaluating and optimising the CNS penetration potential of drug candidates, eventually increasing the likelihood of developing effective therapies for neurological disorders. The E-sol approach developed by Schrodinger’s researchers directly links a compound's 3D structure to its unbound brain-to-plasma partition coefficient (Kp,uu) through a physics-based calculation of solvation free energies. By considering the thermodynamic properties of the molecule in both aqueous and lipid environments, E-sol provides a more accurate and mechanistically driven prediction of brain penetration.
Minimum Solvent-Accessible 3D Polar Surface Area (Min SA 3D PSA): This metric estimates the minimum polar surface area of a molecule accessible to solvent, which correlates with cell permeability. By minimising this surface area, one can potentially enhance a compound's ability to passively diffuse across the BBB.
Quantitative Structure-Activity Relationship (QSAR) Models: We develop and deploy QSAR models using large datasets of known CNS-active compounds and their experimentally determined BBB permeability. By correlating molecular descriptors and fingerprints with BBB penetration, these models can predict the BBB permeability of novel compounds, aiding in lead optimisation and prioritisation. We validate our QSAR models using robust cross-validation techniques and independent test sets to ensure their predictive accuracy and generalisability to new compounds. Cross-validation plays a crucial role in ensuring the robustness and generalisability of our QSAR models for CNS drug discovery. By partitioning the dataset into multiple folds and iteratively training and evaluating the model on different combinations, cross-validation provides a more reliable estimate of the model's predictive performance on unseen data. This technique helps us to avoid overfitting, where the model learns the training data too well but fails to generalise to new compounds. We typically employ k-fold cross-validation, where the dataset is divided into k subsets, and the model is trained and evaluated k times, each time using a different subset as the validation set. This comprehensive validation strategy ensures that our QSAR models are not only accurate but also robust and capable of making reliable predictions, ultimately accelerating the drug discovery process.
For classification models we also assess robustness using metrics such as the Cohen's kappa and the F1-score. The Cohen’s kappa measures agreement between predicted and observed classifications beyond chance, it’s a statistical measure that quantifies the agreement between two raters (or in this case, the model and the true labels) while correcting for the possibility of agreement occurring by chance. It is particularly useful for imbalanced datasets that are quite common in drug discovery. The F1-score is a single score that balances precision (the accuracy of positive predictions) and recall (the ability to find all positive instances). A high F1-score indicates good performance in identifying true positives while minimising false positives and false negatives.
By incorporating these robust validation techniques, we ensure the reliability and predictive power of our machine learning models, further strengthening our ability to accelerate the discovery of effective CNS therapeutics.
Our computational chemistry team is continually exploring innovative approaches to enhance our CNS drug discovery capabilities. We are starting to explore Neural Network-based models (including GNN architectures) that can potentially learn representations of molecular graphs to predict BBB permeability and other CNS-relevant properties. By capturing the complex relationships between atoms and bonds, GNNs offer a flexible framework for modeling complex molecular interactions and predicting drug behavior in the CNS.
We are committed to ongoing research to develop and validate these models, incorporating diverse data sources and exploring new architectures to improve their predictive power. This includes expanding our training datasets with customer’s data (if available) as well as information from the literature. We are also exploring the application of AI-driven generative models to design novel CNS-penetrant molecules with optimised properties.
By combining these cutting-edge approaches with our extensive biosciences, biophysics, medicinal and synthetic chemistry knowledge, we aim to provide our customers with the most advanced techniques for identifying and optimising CNS drug candidates, ultimately accelerating the development of safe and effective therapeutics for neurological disorders.
To find out more and to speak to a member of the computational chemistry team, get in touch.
“The BioAscent in silico discovery team developed a QSAR model based on non-standard activity data that required adaptability and creativity from the BioAscent team. The QSAR model was then used to generate novel compounds and to enumerate focused arrays. Throughout the project, the BioAscent in silico team has been responsive and flexible, which helped us reach a successful conclusion.”
CTO, Tay Therapeutics
Discovery and characterization of novel TRPML1 agonists. X. Peng, C.J. Holler, A-M.F. Alves, M.G. Oliviera, M. Speake, A. Pugliese, M.R. Oskouei, I.D de Freitas, A.Y.-P. Chen, R. Gallegos, S.M. McTighe, G. Koenig, R.S. Hurst, J-F. Blain, J.C. Lanter, D.A. Burnett. View paper.
Implementation of an AI-assisted fragment-generator in an open-source platform. A. E. Bilsland, A. Pugliese, J. Bower. View paper.
Automated Generation of Novel Fragments Using Screening Data, a Dual SMILES Autoencoder, Transfer Learning and Syntax Correction. A. Bilsland, K. McAulay, R. West, A. Pugliese (corresponding author), J. Bower. View paper.
The identification and characterisation of autophagy inhibitors from the published kinase inhibitor sets. M. Zachari, J. Rainard, G. Pandarakalam, L. Robinson, J. Gillespie; M. Rajamanickam; V. Hamon; A. Morrison; I. Ganley; S. McElroy. View paper.
Synthesis and structure–activity relationships of N-(4-benzamidino)-oxazolidinones–potent and selective inhibitors of kallikrein-related peptidase 6; chemRxiv DOI: 10.26434/chemrxiv.9788276, 2019. E. De Vita, N. Smits, H. van den Hurk, E. Beck, J. Hewitt, G. Baillie, E. Russell, A. Pannifer, V. Hamon, A.Morrison, S. McElroy, P. Jones, N. Ignatenko, N. Gunkel and A. Miller. View paper.
Deep generative molecular design – AI at the service of the drug designer. Pharmacology Matters, British Pharmacological Society's online magazine, 2019. A. Pugliese and J. Bower. View paper.
Structure-based design, synthesis and biological evaluation of a novel series of isoquinolone and pyrazolo[4,3-c]pyridine inhibitors of fascin 1 as potential anti-metastatic agents. Bioorg Med Chem Lett. 2019; 29: 1023-9. S. Francis, et al. View paper.
Strategies for Fragment Library Design. In: Erlanson, D.A. and Jahnke W. eds. Fragment-based Drug Discovery: Lessons and Outlook. Wiley,pp. 101-117 2016. J. Bower, A. Pugliese, and M. Drysdale.
Identification of a selective G1-phase benzimidazolone inhibitor by a senescence-targeted virtual screen using artificial neural networks. Neoplasia. 2015 Sep; 17(9):704-15. A. Bilsland, A. Pugliese, et al. View paper.
Fragment-Based Hit Identification – Thinking in 3D. Drug Discov Today. 2013 Dec; 18 (23-24):1221-7. AD. Morley, A. Pugliese, et al. View paper.
Computational tools and resources for metabolism-related property predictions. 2. Application to prediction of half-life time in human liver microsomes. Future Med Chem. 2012 Oct; 4(15):1933-44. AV. Zakharov, ML. Peach, M. Sitzmann, IV. Filippov, HJ. McCartney, LH. Smith, A. Pugliese, MC Nicklaus. View paper.
Software and Resources for Computational Medicinal Chemistry. Future Med Chem. 2011 Jun; 3(8):1057-85. C. Liao, M. Sitzmann, A. Pugliese, M. Nicklaus. View paper.
Posters & case studies
The European Lead Factory: A Collaborative Approach to Drug Discovery. ESC Chemistry: F. MacLeod, A. Morrison, E. Beck, P. Jones, J. Hewitt, M. Matheson, J. Schulz, J. Robinson, L. Robinson, J. Gillespie, M. Huggett, M. Rajamanickam; ESC CMC: A. Pannifer, V. Hamon; ESC Biology: S. McElroy, M. Speake, G. Baillie, E. Russell, J. Rainard, A. Porter, G. Pandarakalam, N. Clark, D. Tegazzini. PPSC: H. Rutjes, C.A.A. van Boeckel, N. Smits, E. van Doornmalen, T. W. Lam, M. Bras, C. S. Schofield; J. J. Brem; M. A. McDonough, M. Van der Stelt, F. J. Janssen, P.P. Geurink, H. Ovaa, H. S. Overkleeft, A. C. M. van Esbroeck, M. P. Baggelaar, H. den Dulk. Download PDF
Harnessing the power of AI to focus compound collections: a collaboration with Medicines Discovery Catapult. Download PDF
