Elix, CBI2026, Oral Presentation, Synthesizer: Directed Chemical Compound Structure Generation and Optimization Driven by Forward Synthesis

Transcript

1. Any unauthorized reproduction of this material is strictly prohibited. Synthesizer: Directed Chemical Compound Structure Generation and Optimization Driven by Forward Synthesis Haris Hasić, Research Engineer @ Elix, Inc. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved
2. Introduction ￭ The chemical compound structures are generated and optimized by assembling predetermined building block chemical compounds according to predetermined chemical reaction templates, establishing the corresponding theoretical chemical synthesis routes in the process. ￭ The building block chemical compounds and chemical reaction templates are predetermined. ? … CC(=O)Nc1ccc (OC(=O)CN(CC)CC)cc1 … CC(=O)Nc1ccc(O)cc1 … CC(=O)Nc1ccccc1 ? CCOc1ccc(NC(C)=O)cc1 ? … Nc1ccccc1 ￭ The chemical compound structures have the corresponding theoretical chemical synthesis routes. ￭ The chemical compound structures are evaluated against a predetermined optimization objective. ￭ The evaluation scores of the chemical compound structures are propagated back, influencing the subsequently utilized building block chemical compounds and chemical reaction templates. Nc1ccc(cc1)S(N)(=O)=O … Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved Evaluation Score 2
3. Advantages and Disadvantages ￭ This approach to generating and optimizing chemical compound structures has the primary advantage of establishing the corresponding theoretical chemical synthesis routes and the primary disadvantage of a limited explorable chemical space. ￭ The explorable chemical space is limited by the building block chemical compounds and chemical reaction templates. – CC(=O)Nc1ccc(OC(=O)CN(C C)CC)cc1 CC(=O)Nc1ccc(O)cc1 … CC(=O)Nc1ccccc1 – CCOc1ccc(NC(C)=O)cc1 + … Nc1ccccc1 ￭ The chemical compound structures have the corresponding theoretical chemical synthesis routes. Nc1ccc(cc1)S(N)(=O)=O … ￭ The process of forward synthesis is primarily additive, resulting in chemical compound structures outside of the relevant chemical space after a couple of iterations. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 3
4. Objectives ￭ This approach to generating and optimizing chemical compound structures can be improved by controlling the selection of chemical compounds and the application of chemical reaction templates. ￭ Controlling the chemical compound selection and chemical reaction template application reduces the explorable chemical space. ! CC(=O)Nc1ccc(OC(=O)CN(C C)CC)cc1 CC(=O)Nc1ccc(O)cc1 … CC(=O)Nc1ccccc1 ! CCOc1ccc(NC(C)=O)cc1 … Nc1ccccc1 Nc1ccc(cc1)S(N)(=O)=O … ￭ Controlling the chemical compound selection and chemical reaction template application enables a more detailed exploration of the relevant chemical space. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 4
5. Elix, Inc. Synthesizer ￭ The name of the internal approach to directed chemical compound structure generation and optimization driven by forward synthesis by Elix, Inc. Synthesizer and it consists of four primary components: curation, transformation, evaluation, and prospection. ￭ The approach is intentionally modular and compatible with the Elix, Inc. Discovery™ platform. Elix, Inc. Synthesizer Components Curation Transformation Evaluation Prospection ￭ The curation component ingests and indexes the building block chemical compounds, chemical reaction templates, and chemical synthesis route seeds. ￭ The transformation component applies the chemical reaction templates to reactant compounds, generating product compounds in the process. ￭ The evaluation component scores and filters the chemical compound structures against a predetermined optimization objective. ￭ The prospection component explores the relevant chemical space and manages the generation and optimization of chemical compound structures. ￭ The chemical synthesis route seed adds direction to the generation and optimization process by constraining the utilized chemical compound structures and reaction templates. ￭ Currently, the implemented backends are RDKit and RDChiral. ￭ Currently, the implemented backend is Discovery. ￭ Currently, the implemented backends are Monte Carlo Tree Search (MCTS), and A* Search. ￭ Currently, the implemented backends are Random and Informed. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 5
6. Chemical Synthesis Route Seed ￭ The chemical synthesis route seed adds direction to the generation and optimization process by constraining the utilized building block, intermediate, and target chemical compounds and reaction templates. { Nc1ccccc1 "target_compound_frozen_substructure_templates": [ { "smarts": "c1ccccc1", "is_accumulating": false } ], "reactions": { "0": { "included_templates": { "smarts_or_tag_strings": [ "amide_coupling" ], "is_hard_constrained": true }, "reactant_compounds": { "1": { "secondary": { "compound": { "smiles": "CC(=O)OC(C)=O", "is_hard_constrained": true } } } } } } CC(=O)OC(C)=O CC(=O)Nc1ccccc1 CC(=O)Nc1ccc(O)cc1 OC(=O)CN(CC)CC Chemical Synthesis Route Seed ￭ The chemical synthesis route seed supports freezing, inclusion, exclusion, and reference for chemical compounds and reaction templates as well as their corresponding tags. ￭ The constraints can be soft or hard, depending on how flexible the exploration of the chemical space should be. } CC(=O)Nc1ccc(OC(=O)CN(CC)CC)cc1 Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 6
7. Evolving Chemical Synthesis Route ￭ The chemical compound structure generation and optimization process can be extended over multiple rounds, utilizing beam search to evolve chemical synthesis route seeds. ￭ Each round is summarized through a corresponding chemical synthesis route seed alongside a the generated and optimized chemical compound structures. ￭ The actor in control of the chemical synthesis route seed evolution can be the algorithm, an agent, or a human. B Synthesizer Algorithm A Synthesizer AI Agent E … F X G … H … I X Algorithm C X AI Agent Human Human D Synthesizer Chemical Compound Structures Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 7
8. Initial Benchmarking Results ￭ Hourly Turnaround → Depending on the evaluation settings, a single round of directed generation and optimization of the chemical compound structures can be completed within hours, which is performant enough to run multiple rounds within a single day. ￭ Valid Chemistry → Every chemical compound structure is computed and not predicted, which ensures the chemical validity of the structures renders invalidity or hallucinations impossible. ￭ Real World Use Cases → The quantitative benchmarks require concrete real world chemical compound structure targets and objectives, making the application of the approach to practical use cases the immediate next step. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 8
9. Conclusion and Future Work ￭ The Elix, Inc. Synthesizer approach generates and optimizes chemical compound structures driven by forward synthesis. ￭ The chemical synthesis route seed adds direction and control, something that the majority of approaches from this category lack. ￭ The evolution of chemical synthesis route seeds scales a single round of generation and optimization into a beam search over multiple rounds, steered by a pluggable actor with a global pool of chemical compound structures. ￭ Even at an early stage of development and testing, the approach is performant in both the execution speed and chemistry dimension. ￭ The primary challenge in the future, however, are concrete drug discovery use cases accompanied by corresponding quantitative benchmarks. Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved 9
10. Thank You for Your Attention !!! Chem-Bio Informatics Society (CBI) Annual Meeting 2026, Tokyo, Japan | October 29th, 2026 | Copyright © Elix, Inc. All Rights Reserved