The paper presents a novel approach to chemical synthesis planning using Monte Carlo Tree Search (MCTS) combined with deep neural networks. The authors aim to automate the process of retrosynthetic analysis, which is a crucial technique for planning the synthesis of small organic molecules. Traditional methods have been slow and often produce unsatisfactory results. By training deep neural networks on a large dataset of reactions, the system can efficiently discover retrosynthetic routes. The neural networks include an expansion policy network that guides the search, an in-scope filter network to pre-select promising steps, and a rollout policy network to estimate position values. The 3N-MCTS method outperforms traditional search methods in terms of speed and the number of solved problems. Double-blind AB tests with organic chemists show that the proposed method produces more chemically reasonable routes compared to literature routes and traditional heuristics. The authors conclude that their approach can significantly accelerate drug and materials discovery by assisting chemists in planning better syntheses faster and enabling fully automated robot synthesis.The paper presents a novel approach to chemical synthesis planning using Monte Carlo Tree Search (MCTS) combined with deep neural networks. The authors aim to automate the process of retrosynthetic analysis, which is a crucial technique for planning the synthesis of small organic molecules. Traditional methods have been slow and often produce unsatisfactory results. By training deep neural networks on a large dataset of reactions, the system can efficiently discover retrosynthetic routes. The neural networks include an expansion policy network that guides the search, an in-scope filter network to pre-select promising steps, and a rollout policy network to estimate position values. The 3N-MCTS method outperforms traditional search methods in terms of speed and the number of solved problems. Double-blind AB tests with organic chemists show that the proposed method produces more chemically reasonable routes compared to literature routes and traditional heuristics. The authors conclude that their approach can significantly accelerate drug and materials discovery by assisting chemists in planning better syntheses faster and enabling fully automated robot synthesis.