TY - JOUR T1 - Artificial Intelligence in Nuclear Medicine for Brain Imaging. JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3016 LP - 3016 VL - 62 IS - supplement 1 AU - Ming-Hsin Li AU - Kai-Hung Cheng AU - Shih-Wei Lo AU - Liang-Hsun Huang AU - Li-Ming Wang Y1 - 2021/05/01 UR - http://jnm.snmjournals.org/content/62/supplement_1/3016.abstract N2 - 3016Introduction: The vast medical chemical space (comprising >1060 molecules) fosters the development of a large number of drug molecules. Artificial intelligence (AI) - that is including both deep learning and machine intelligence - has the potential to make waves in drug discovery. In the past, researchers have attempted to develop expert systems and knowledge-reasoning systems to solve problems in medical chemistry. Because AI is developing quickly that is therefore timely that we consider what is involved and how we should prepare if we are to attach the competitive advantages of nuclear medicine.However, it is not without its challenges. In this issue of Nuclear Medicine for Brain Imaging, we focus on AI for small-molecule-drug discovery purposes, and in this study we take a look at how AI is being utilized throughout the drug design and development processes, and the positives and negatives posed by our AI technology.Our Variational Auto Encoder (VAE) algorithm uses a python-based AI system to find a suitable candidate in drug discovery. The Figure shows VAE architecture for our brain imaging AI model.VAE starts by encoding the molecule in SMILES(simplified molecular-input line-entry system) string using the encoder function, then applying convolutions over various filter sizes that correspond to chemical substructures. During generation, the vectors of features are sampled from the prior distribution and their output is passed to the decoder that generates a new representation.This VAE model had trained on a subset with approximately 250k drug-like compounds extracted from ZINC database (a free tool to discover chemistry for biology) randomly. In order to make the resulting compounds more novel, we had added a layer of diversity before the decoding layer to produce similar but different structures. In the end of training, our model has reached 97.68% accuracy. In the forecasting stage, we has used MK-6240 analogue of 28 tau imaging candidates (from MK-6240 patent own by Merck.com) as input for novel drug prediction, proposing testable hypotheses. It is expected that generated drugs should be closed to original compounds and have similar medicinal properties to prove hypotheses for both AI and chemistry. We measure our generated drugs with the following metrics: valid compound and novelty. The result shows we has generated 25/28(89.2% odds ratio) valid and novel drug-like molecules.Most importantly, the development of one new drug typically takes over 10 years and costs over one billion US dollars. For saving the huge cost, we generated a creatively think of new hypotheses about drug discovery. The next step, our chemist needs to experimentally validate each novel drug-like molecule and the additional requirements like efficacy and non-toxicity are fulfilled. We will use the gained insights to validate new AI knowledge and creatively think of new hypotheses about drug discovery and development for brain imaging.$$graphic_{1EAE5047-4448-41BC-8F59-66C1ED08E2E3}$$ ER -