School of Science Department of Chemistry 6 Deep Learning in Synthesis Planning Supervisor: SU Haibin / CHEM Student: CHIU King Wai / CHEM Course: UROP1100, Spring Machine learning in Chemistry becomes hot topic currently. Computer-aided synthesis planning (CASP) with retrosynthesis can be constructed for chemists to search for plausible most effective and cheapest synthesis route with machine learning technique, such as Artificial Neural Networks (ANN) and Monte Carlo Tree Search (MCTS). However, commercial database like Reaxys cannot provide accurate and detailed chemicals and reaction condition, especially when organometallic reaction is considered. In this project, automatic data-mining tools are constructed with Optical Structure Recognition Software (OSRA) and Optical Character Recognition (OCR) Software which utilized Long Short-Term Memory (LSTM) model. In future, there is high potential to build a machine learning model in chemistry paper reading to further improve the data-mining. Important reaction elements, such as catalysts, ligands, additives, temperature and yield can be extracted to enrich the database to outperform the current available commercial and open source and be treated as more detailed descriptors to feature the reaction rules for better retrosynthesis planning by SPLASH. Deep Learning in Synthesis Planning Supervisor: SU Haibin / CHEM Student: CHUI Sin Yu / CHEM Course: UROP1100, Spring Conditions of reactions are important to determine and proceed synthesis planning that is comprehensively applied in many fields of chemistry. Reaction data analysis may help chemists to make decisions whether the pathway would be high-efficient without on-hands laboratory work. Utilizing which type of catalyst is also one critical condition required in catalytic reactions for organic synthesis. This report is an overview of reactions using nickel catalyst with bidentate nitrogen ligands by reaction data analyzing. Nine types of reactions are covered including Suzuki coupling, Negishi coupling, Hiyama reaction and C-H activation with the usage of 21 ligands like BBBPY, bpy, BPhen and (R,R)-ph-box. Deep Learning in Synthesis Planning Supervisor: SU Haibin / CHEM Student: FUNG Ka Shing / DSCT Course: UROP1000, Summer Retrosynthesis is one of the hottest topics in chemistry. Our aim is to build a retrosynthesis program with the help of machine learning. SMILES and SMARTS, the notation of compounds briefly explained in computer. Monte Carlo Tree Search and artificial neural network, computer software RDChiral, Reaction Decoder are applied. A neural network was trained and verified by testing. Two compounds, curcumin and vortioxetine were put into the model, successfully generated some positive results. There were a few failure cases due to reaction centre selectivity and ring closing problem. Further improvements could be made by importing more conditions into the neural network and tree searching criteria.