Positron emission tomography (PET) depends on small-molecule radioligands that bind a chosen biological target with high affinity and selectivity while crossing the blood-brain barrier and showing low nonspecific binding. Developing such ligands for central nervous system (CNS) targets remains slow and attrition-prone, particularly for biomarkers relevant to Alzheimer's disease. This project will apply state-of-the-art generative machine learning algorithms to design novel small-molecule ligands tailored to the demanding property profile of CNS PET radioligands, with the aim of accelerating tracer and lead discovery for neurodegeneration in vivo biomarker research. The computational approach combines two complementary deep-learning methods. REINVENT4, an open-source reinforcement-learning framework for de novo molecular generation, is run in iterative optimisation loops against multi-parameter objectives encoding predicted affinity, selectivity, synthesisability, lipophilicity, brain penetration and suitability for carbon-11 or fluorine-18 labelling. Generated candidates are then evaluated with Boltz-2, a deep-learning co-folding and binding-affinity model, to predict ligand-target complexes and rank molecules by structural plausibility and predicted binding, providing structure-based feedback even where experimental structures are limited. Lightweight cheminformatics and in silico ADMET filters, triage the output into a ranked, synthetically tractable shortlist for subsequent synthesis, radiolabelling and experimental validation. Both core methods are GPU-bound and constitute the dominant computational cost. REINVENT4 optimisation campaigns require many parallel runs across scoring configurations and restarts, while Boltz-2 complex and affinity prediction is GPU- and memory-intensive per evaluation and is applied across large generated candidate sets. The iteration depth needed for a realistic ligand-design campaign exceeds locally available hardware. Access to a NAISS GPU resource would enable these workflows at the scale required to deliver credible candidate series, supporting the research programme for molecular imaging in neurodegeneration.