I’ll cover a framework we call ACD (Automated Chemical Design) Levels for describing the level of autonomy for AI-powered systems that design molecules. This framework allows relevant distinctions to be drawn and can help teams better understand the claimed capabilities of a system and ask more insightful questions to understand their function. Using these definitions, we will discuss the challenges and opportunity for AI-powered design of molecules in drug discovery.

Generative Adversarial Networks (GANs) and Reinforcement Learning (RL) have been successfully applied for de novo drug design. Multiple such frameworks have been explored and shared by many AI researchers. We review these models and discuss the features of these techniques in terms of generating novel molecules with desired properties.

The Critical Assessment of Computational Hit-Finding Experiments (CACHE) is an international competition where computational chemists and AI experts around the world submit their predicted compounds which are then procured and rigorously tested experimentally at CACHE against a pre-defined target. The first two challenges, to find hits for the Parkinson’s disease target LRRK2 and the SARS-CoV-2 helicase NSP13 are ongoing. ~50% of participants combined physics-based and deep-learning techniques. Do these hybrid methods already out-perform more traditional approaches?

Solubility can be a challenging, time-consuming, material intensive property to measure correctly. To accurately predict solubility values of drug-like compounds, we have recently developed a molecular featurization scheme – MEMS, or Manifold Embedding of Molecular Surface – based on electronic properties on the surface of a molecule. Deep learning methods were further developed. Our effort shows promise in solubility prediction with an accuracy outperforming most of the reported models.

Exscientia’s patient-first approach to drug discovery has produced industry-leading productivity. Structural information is crucial across this process – from target ID, hit generation, and compound optimisation. The revolution in availability of computational and experimental structural information further drives Exscientia's efficiency and predictive quality. In this talk, we outline how we leverage the power of physics and informatics in our AI algorithms. We exemplify this through high-throughput structural target assessment, delivery of novel and class-leading potential anti-viral programme for SARS-CoV-2, and in delivering automated and de novo potent hits in the first cycle of design for a kinase.

We will showcase how data analytics/AI is being used in business processes and how this improves decision-making for drug design teams. We will present retro- and prospective studies to highlight the value proposition of the innovation integration of novel science and technology into business processes. A key differentiator for enabling analytics/AI is to decide which science and technology contributes to the business and only lift those to enterprise level.

BigHat Biosciences is designing safer, more effective antibody therapies for patients using machine learning and synthetic biology. Machine learning guides the search for better molecules by directing and learning from each cycle of our high speed, automated wet lab that synthesizes and characterizes hundreds of antibodies each week. We’ll highlight key features of our platform and share several case studies of protein engineering using this novel platform.

We developed a method for de novo antibody design using antigen structure alone. Across multiple antigens, we characterized binding strength for ~10⁵ antibody designs and observed 10s-100s of strong binders. Further, all designs were assayed in multiplex for stability and polyspecificity, and this revealed a broad array of binding and developability trade-offs. These results highlight the importance of integrating machine learning with high-throughput, multi-property measurements for the holistic design of antibody therapeutics.

Successful peptide drug discovery programs today require attainment of multiple performance metrics to progress a compound to clinical stage. To aid decision-making, Koliber has developed an AI peptide platform based on state-of-the-art machine learning methods to analyze peptide properties, profile positions, and predict new variants. The capabilities and wet-lab validation of the AI platform will be demonstrated with examples from immunology and antimicrobial peptide discovery and optimization.