AI/ML for Early Drug Discovery

The evolution of drug discovery is increasingly driven by remarkable advances in AI, simulation, and data integration technologies. SandboxAQ generates proprietary data using physics-based methods, and trains Large Quantitative Models (LQMs). We will highlight and present the application of our methods for hit finding and lead optimisation applied to promising drug discovery targets for neurodegeneration.

Co-folding has emerged as a tool holding promise to revolutionise drug discovery. Beyond the prediction of protein-ligand binding modes, co-folding can be extended to tackle multiple tasks in drug discovery. In this presentation I will discuss the benchmarking of co-folding in several drug discovery applications and a success case on the identification of a cryptic pocket.

As generative models grow more complex, discerning their actual contributions to drug design becomes challenging. This presentation assesses the impact of these models, focusing on molecule synthesisability and 3D integration. We critically analyse limitations from small datasets and the models' tendency to infer patterns without genuine extrapolative power. Emphasising need for clarity in evaluation, we propose strategies for meaningful benchmarks to ensure generative models deliver tangible improvements in drug discovery.

Structure-based generative modelling (SBGM) represents a change of paradigm in drug discovery, from virtually screening ultra-large chemical libraries to virtually growing molecules with desired physicochemical and ADME properties directly inside the protein cavity. In this talk, the SBGM platform developed at Chemotargets to generate novel synthetically feasible drug-like molecules for protein targets will be introduced. Examples of fragment evolution inside protein cavities will be presented.

PlayMolecule AI is an AI co-scientist platform designed to accelerate discovery in drug research by integrating natural language querying, multidimensional data analysis, and interactive 3D molecular visualisation. Researchers can interrogate chemical and biological datasets, explore molecular interactions, and perform advanced computational experiments, including molecular dynamics simulations, virtual screening, and property prediction, all within a unified environment. Built on scientifically validated computational workflows rigorously tested both internally and through collaboration, the platform empowers scientists to unlock their full potential without concerns about gaps in expertise and enables rapid iteration of validation tests.

In this talk we will summarize recent and novel research highlighting the use of active learning cycles in hit finding, the development and use of libraries of 100s of billions of molecules and exploring AI methods for advancing polypharmacology.

The Expert Systems AIML platform provides simultaneous predictions for thousands of machine learning models. Here, we briefly introduce the ExSys AIML platform, focusing on prediction prioritisation. Given appropriate context, our platform quickly sifts through all the predictions, eliminates those that do not match predefined criteria (e.g., prediction quality, model size), provides a qualitative assessment, and highlights predictions that require further action. This post-ML system is designed to quickly sift through thousands of predictions and flag both problematic compounds and those with promising potential.

Kuano’s proprietary combination of advanced quantum methods enables sophisticated and highly-accurate predictions of complex systems, including transition states and transition metals. ‘Quantum Pharmacophores’ coupled with AI-assisted workflow guide the design of small molecule inhibitors with enhanced specificity, binding complementarity, and potential for covalency. Pipeline examples presented include: in vivo efficacy for a first-in-class colorectal cancer target; resolving selectivity-associated toxicity for methyltransferase DNMT1; selective phosphatase inhibitors; and covalent reactivity prediction.