Jihye Roh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jihye Roh has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Connor W. Coley
Massachusetts Institute of Technology
8 shared publications Zhengkai Tu
Department of Electrical Engineering and Computer Science
5 shared publications- Joonyoung Francis Joung
Department of Chemistry Kookmin University Seoul Korea
3 shared publications Mun Hong Fong
Department of Chemical Engineering
3 shared publicationsJohn Bradshaw
Department of Chemical Engineering Massachusetts Institute of Technology, Cambridge Massachusetts 02139 United States
2 shared publicationsKevin Yu
Center for Computational Science and Engineering
2 shared publications
Latest Publications
- 24 June 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (13) , 7489-7500
Pathway-Aware Template-Based Retrosynthesis
There has been growing interest in developing machine-learning retrosynthesis models to accelerate chemical synthesis, enabling the discovery of routes to synthesize high-value products such as small-molecule drugs. However, most single-step models within a multistep planning algorithm perform recursive predictions without considering the previous reaction steps in the retrosynthetic pathway, which may lead to inefficiencies in the search, as these previous steps can provide important context for deciding which reactions to follow. We introduce the PATRO (Pathway-Aware Template-based RetrOsynthesis) model, which augments a template-based single-step retrosynthesis model by processing pathway-level information with a Long Short-Term Memory (LSTM). Compared to the baseline model, we demonstrate improvements in both single-step and multistep retrosynthesis, which can be attributed to the incorporation of pathway-level information and related architectural modifications. In single-step retrosynthesis, PATRO outperformed the baseline by 2.3% in top-1 accuracy, demonstrating improved template predictions when considering pathway context. After integrating the pathway-aware model as the expansion policy for two multistep retrosynthesis algorithms, we demonstrate that PATRO provides consistent performance gains over the baseline model across multiple multistep metrics, including success rate, patent route recovery rate, and top-k accuracy. The PATRO model also enables more efficient planning to discover the literature routes extracted from patents, requiring on average 10% fewer iterations. While demonstrated here within a well-defined template-based framework, we believe that this strategy of incorporating pathway-level information could benefit other diverse retrosynthesis models.
- 24 June 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 66 (13) , 7489-7500
Pathway-Aware Template-Based Retrosynthesis
There has been growing interest in developing machine-learning retrosynthesis models to accelerate chemical synthesis, enabling the discovery of routes to synthesize high-value products such as small-molecule drugs. However, most single-step models within a multistep planning algorithm perform recursive predictions without considering the previous reaction steps in the retrosynthetic pathway, which may lead to inefficiencies in the search, as these previous steps can provide important context for deciding which reactions to follow. We introduce the PATRO (Pathway-Aware Template-based RetrOsynthesis) model, which augments a template-based single-step retrosynthesis model by processing pathway-level information with a Long Short-Term Memory (LSTM). Compared to the baseline model, we demonstrate improvements in both single-step and multistep retrosynthesis, which can be attributed to the incorporation of pathway-level information and related architectural modifications. In single-step retrosynthesis, PATRO outperformed the baseline by 2.3% in top-1 accuracy, demonstrating improved template predictions when considering pathway context. After integrating the pathway-aware model as the expansion policy for two multistep retrosynthesis algorithms, we demonstrate that PATRO provides consistent performance gains over the baseline model across multiple multistep metrics, including success rate, patent route recovery rate, and top-k accuracy. The PATRO model also enables more efficient planning to discover the literature routes extracted from patents, requiring on average 10% fewer iterations. While demonstrated here within a well-defined template-based framework, we believe that this strategy of incorporating pathway-level information could benefit other diverse retrosynthesis models.
- Open Access
- 4 June 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 17 (28) , 14054-14063
Quantifying the failure modes of current one-step retrosynthesis models
Computer-aided synthesis planning (CASP) automates retrosynthetic analysis, generally by recursively applying one-step retrosynthesis models within multistep search algorithms to simplify a target molecule into commercially available starting materials. Despite their utility, these tools often fail to recover literature-reported pathways. Such failures arise from two causes: either (i) the literature-reported precursor is not proposed at all or (ii) it is proposed but ranked too low to be discovered during a multistep search. In this work, we quantify the challenges that data-driven one-step retrosynthesis models face in reproducing literature-reported precursors. We first evaluate model performance using standard top-k exact-match accuracy and stratify this accuracy by product and reaction complexity, demonstrating a decrease in performance with increasing complexity. This decline is accompanied by a systematic underprediction of the number of reacting atoms and changing rings, indicating a bias toward simpler transformations, even when complex examples are included in the training data. To gain deeper insights into failure modes, we evaluate models with complementary metrics that account for incorrect stereochemistry, leaving groups, and multi-stage reactions. Overall, our work provides a quantitative analysis of how one-step retrosynthesis models fail to capture literature-reported reactions, highlighting opportunities for improving future models and providing guidance on using model predictions more effectively in prospective synthesis planning.
- 4 June 2026
- Published by Royal Society of Chemistry (RSC) in Chemical Science
- Vol. 17 (28) , 14054-14063
Open AccessQuantifying the failure modes of current one-step retrosynthesis models
Computer-aided synthesis planning (CASP) automates retrosynthetic analysis, generally by recursively applying one-step retrosynthesis models within multistep search algorithms to simplify a target molecule into commercially available starting materials. Despite their utility, these tools often fail to recover literature-reported pathways. Such failures arise from two causes: either (i) the literature-reported precursor is not proposed at all or (ii) it is proposed but ranked too low to be discovered during a multistep search. In this work, we quantify the challenges that data-driven one-step retrosynthesis models face in reproducing literature-reported precursors. We first evaluate model performance using standard top-k exact-match accuracy and stratify this accuracy by product and reaction complexity, demonstrating a decrease in performance with increasing complexity. This decline is accompanied by a systematic underprediction of the number of reacting atoms and changing rings, indicating a bias toward simpler transformations, even when complex examples are included in the training data. To gain deeper insights into failure modes, we evaluate models with complementary metrics that account for incorrect stereochemistry, leaving groups, and multi-stage reactions. Overall, our work provides a quantitative analysis of how one-step retrosynthesis models fail to capture literature-reported reactions, highlighting opportunities for improving future models and providing guidance on using model predictions more effectively in prospective synthesis planning.
- Top Cited PapersOpen Access
- 5 March 2026
- Published by American Chemical Society (ACS) in ACS Central Science
- Vol. 12 (3) , 345-357
Higher-Level Strategies for Computer-Aided Retrosynthesis
Retrosynthesis is a core technique in organic chemistry that simplifies target molecules into more readily available components. Computer-aided synthesis planning (CASP) automates this process by recursively proposing immediate precursors to identify multistep synthetic pathways. However, CASP typically struggles for complex molecules that require longer synthetic pathways and present a greater number of possible disconnections. Here, we introduce a new higher-level framework for computer-aided retrosynthesis. Our approach abstracts detailed substructures in pathway intermediates not appearing in the target product, allowing the algorithm to emphasize higher-level strategies while postponing the consideration of specific functional group choices, thus reducing the effective width and depth of the search space. This framework achieves higher top-k accuracy in single-step retrosynthesis and identifies multistep routes for more targets than the original approach. Through case studies on complex drugs and natural products, we demonstrate how routes proposed by our framework provide a powerful basis for developing full synthesis plans, particularly in challenging cases where the original approach fails, while enabling chemists to leverage their expertise to refine the synthesis design. Ultimately, focusing on higher-level strategies enables an effective and intuitive approach for challenging targets in computer-aided retrosynthesis.
- 5 March 2026
- Published by American Chemical Society (ACS) in ACS Central Science
- Vol. 12 (3) , 345-357
Top Cited PapersOpen AccessHigher-Level Strategies for Computer-Aided Retrosynthesis
Retrosynthesis is a core technique in organic chemistry that simplifies target molecules into more readily available components. Computer-aided synthesis planning (CASP) automates this process by recursively proposing immediate precursors to identify multistep synthetic pathways. However, CASP typically struggles for complex molecules that require longer synthetic pathways and present a greater number of possible disconnections. Here, we introduce a new higher-level framework for computer-aided retrosynthesis. Our approach abstracts detailed substructures in pathway intermediates not appearing in the target product, allowing the algorithm to emphasize higher-level strategies while postponing the consideration of specific functional group choices, thus reducing the effective width and depth of the search space. This framework achieves higher top-k accuracy in single-step retrosynthesis and identifies multistep routes for more targets than the original approach. Through case studies on complex drugs and natural products, we demonstrate how routes proposed by our framework provide a powerful basis for developing full synthesis plans, particularly in challenging cases where the original approach fails, while enabling chemists to leverage their expertise to refine the synthesis design. Ultimately, focusing on higher-level strategies enables an effective and intuitive approach for challenging targets in computer-aided retrosynthesis.
- Open Access
- 15 February 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Quantifying the Failure Modes of Current One-step Retrosynthesis Models
Computer-aided synthesis planning (CASP) automates retrosynthetic analysis, generally by recursively applying one-step retrosynthesis models within multistep search algorithms to simplify a target molecule into commercially available starting materials. Despite their utility, these tools often fail to recover literature-reported pathways. Such failures arise from two causes: either (i) the literature-reported precursor is not proposed at all or (ii) it is proposed but ranked too low to be discovered during multistep search. In this work, we quantify the challenges that data-driven one-step retrosynthesis models face in reproducing the reported precursors. We first evaluate model performance using standard top- k exact-match accuracy and stratify this accuracy by product and reaction complexity, demonstrating a decrease in performance with increasing complexity. This decline is accompanied by a systematic underprediction of the number of reacting atoms and changing rings, indicating a bias toward simpler transformations, even when complex examples are included in the training data. To gain deeper insights into failure modes, we evaluate models with complementary metrics that account for incorrect stereochemistry, leaving groups, and multi-stage reactions. Overall, our work provides a quantitative analysis of how one-step retrosynthesis models fail to capture literature-reported reactions and highlights opportunities for improving future models.
- 15 February 2026
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessQuantifying the Failure Modes of Current One-step Retrosynthesis Models
Computer-aided synthesis planning (CASP) automates retrosynthetic analysis, generally by recursively applying one-step retrosynthesis models within multistep search algorithms to simplify a target molecule into commercially available starting materials. Despite their utility, these tools often fail to recover literature-reported pathways. Such failures arise from two causes: either (i) the literature-reported precursor is not proposed at all or (ii) it is proposed but ranked too low to be discovered during multistep search. In this work, we quantify the challenges that data-driven one-step retrosynthesis models face in reproducing the reported precursors. We first evaluate model performance using standard top- k exact-match accuracy and stratify this accuracy by product and reaction complexity, demonstrating a decrease in performance with increasing complexity. This decline is accompanied by a systematic underprediction of the number of reacting atoms and changing rings, indicating a bias toward simpler transformations, even when complex examples are included in the training data. To gain deeper insights into failure modes, we evaluate models with complementary metrics that account for incorrect stereochemistry, leaving groups, and multi-stage reactions. Overall, our work provides a quantitative analysis of how one-step retrosynthesis models fail to capture literature-reported reactions and highlights opportunities for improving future models.
- Open Access
- 21 December 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Higher-level Strategies for Computer-Aided Retrosynthesis
Retrosynthesis is a core technique in organic chemistry that involves simplifying target molecules into more readily available components. Computer-aided synthesis planning (CASP) tools automate this process by recursively proposing immediate precursors of molecules to identify multistep synthetic pathways. However, the application of CASP tools becomes particularly challenging for complex molecules due to the increased length of synthetic pathways and the increased number of possible disconnections. We introduce a new framework for computer-aided retrosynthesis, \emph{higher-level} retrosynthesis, that focuses on the broader retrosynthetic strategy. We abstract the detailed substructures in the intermediates of a synthetic pathway that do not appear in the target product, allowing our algorithm to emphasize the general, higher-level strategies and postpone the consideration of chemically-equivalent functional groups to a later stage. This approach leads to shorter and narrower search trees, letting the search algorithm allocate its computational resources more efficiently and discover retrosynthetic strategies that the original approach does not generate. We demonstrate the empirical effectiveness of our approach with higher success rates in single- and multi-step retrosynthetic planning. We further underscore the value of our framework through case studies involving complex drugs and natural products. We illustrate how the strategies identified through our framework provide chemists with a powerful basis to develop full synthesis plans, particularly in challenging cases where the original approach to computer-aided retrosynthesis trained on the same dataset yields no viable solutions, while also enabling them to leverage their expertise and intuition to refine the synthesis design process. Ultimately, focusing on higher-level strategies in computer-aided retrosynthesis enhances its effectiveness and enables a more intuitive approach to addressing challenging retrosynthesis problems.
- 21 December 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessHigher-level Strategies for Computer-Aided Retrosynthesis
Retrosynthesis is a core technique in organic chemistry that involves simplifying target molecules into more readily available components. Computer-aided synthesis planning (CASP) tools automate this process by recursively proposing immediate precursors of molecules to identify multistep synthetic pathways. However, the application of CASP tools becomes particularly challenging for complex molecules due to the increased length of synthetic pathways and the increased number of possible disconnections. We introduce a new framework for computer-aided retrosynthesis, \emph{higher-level} retrosynthesis, that focuses on the broader retrosynthetic strategy. We abstract the detailed substructures in the intermediates of a synthetic pathway that do not appear in the target product, allowing our algorithm to emphasize the general, higher-level strategies and postpone the consideration of chemically-equivalent functional groups to a later stage. This approach leads to shorter and narrower search trees, letting the search algorithm allocate its computational resources more efficiently and discover retrosynthetic strategies that the original approach does not generate. We demonstrate the empirical effectiveness of our approach with higher success rates in single- and multi-step retrosynthetic planning. We further underscore the value of our framework through case studies involving complex drugs and natural products. We illustrate how the strategies identified through our framework provide chemists with a powerful basis to develop full synthesis plans, particularly in challenging cases where the original approach to computer-aided retrosynthesis trained on the same dataset yields no viable solutions, while also enabling them to leverage their expertise and intuition to refine the synthesis design process. Ultimately, focusing on higher-level strategies in computer-aided retrosynthesis enhances its effectiveness and enables a more intuitive approach to addressing challenging retrosynthesis problems.
- Top Cited PapersOpen Access
- 21 May 2025
- Published by American Chemical Society (ACS) in Accounts of Chemical Research
- Vol. 58 (11) , 1764-1775
ASKCOS: Open-Source, Data-Driven Synthesis Planning
The advancement of machine learning and the availability of large-scale reaction datasets have accelerated the development of data-driven models for computer-aided synthesis planning (CASP) in the past decade. In this Account, we describe the range of data-driven methods and models that have been incorporated into the newest version of ASKCOS, an open-source software suite for synthesis planning that we have been developing since 2016. This ongoing effort has been driven by the importance of bridging the gap between research and development, making research advances available through a freely available practical tool. ASKCOS integrates modules for retrosynthetic planning, modules for complementary capabilities of condition prediction and reaction product prediction, and several supplementary modules and utilities with various roles in synthesis planning. For retrosynthetic planning, we have developed an Interactive Path Planner (IPP) for user-guided search as well as a Tree Builder for automatic planning with two well-known tree search algorithms, Monte Carlo Tree Search (MCTS) and Retro*. Four one-step retrosynthesis models covering template-based and template-free strategies form the basis of retrosynthetic predictions and can be used simultaneously to combine their advantages and propose diverse suggestions. Strategies for assessing the feasibility of proposed reaction steps and evaluating the full pathways are built on top of several pioneering efforts that we have made in the subtasks of reaction condition recommendation, pathway scoring and clustering, and the prediction of reaction outcomes including the major product, impurities, site selectivity, and regioselectivity. In addition, we have also developed auxiliary capabilities in ASKCOS based on our past and ongoing work for solubility prediction and quantum mechanical descriptor prediction, which can provide more insight into the suitability of proposed reaction solvents or the hypothetical selectivity of desired transformations. For each of these capabilities, we highlight its relevance in the context of synthesis planning and present a comprehensive overview of how it is built on top of not only our work but also of other recent advancements in the field. We also describe in detail how chemists can easily interact with these capabilities via user-friendly interfaces. ASKCOS has assisted hundreds of medicinal, synthetic, and process chemists in their day-to-day tasks by complementing expert decision making and route ideation. It is our belief that CASP tools are an important part of modern chemistry research and offer ever-increasing utility and accessibility.
- 21 May 2025
- Published by American Chemical Society (ACS) in Accounts of Chemical Research
- Vol. 58 (11) , 1764-1775
Top Cited PapersOpen AccessASKCOS: Open-Source, Data-Driven Synthesis Planning
The advancement of machine learning and the availability of large-scale reaction datasets have accelerated the development of data-driven models for computer-aided synthesis planning (CASP) in the past decade. In this Account, we describe the range of data-driven methods and models that have been incorporated into the newest version of ASKCOS, an open-source software suite for synthesis planning that we have been developing since 2016. This ongoing effort has been driven by the importance of bridging the gap between research and development, making research advances available through a freely available practical tool. ASKCOS integrates modules for retrosynthetic planning, modules for complementary capabilities of condition prediction and reaction product prediction, and several supplementary modules and utilities with various roles in synthesis planning. For retrosynthetic planning, we have developed an Interactive Path Planner (IPP) for user-guided search as well as a Tree Builder for automatic planning with two well-known tree search algorithms, Monte Carlo Tree Search (MCTS) and Retro*. Four one-step retrosynthesis models covering template-based and template-free strategies form the basis of retrosynthetic predictions and can be used simultaneously to combine their advantages and propose diverse suggestions. Strategies for assessing the feasibility of proposed reaction steps and evaluating the full pathways are built on top of several pioneering efforts that we have made in the subtasks of reaction condition recommendation, pathway scoring and clustering, and the prediction of reaction outcomes including the major product, impurities, site selectivity, and regioselectivity. In addition, we have also developed auxiliary capabilities in ASKCOS based on our past and ongoing work for solubility prediction and quantum mechanical descriptor prediction, which can provide more insight into the suitability of proposed reaction solvents or the hypothetical selectivity of desired transformations. For each of these capabilities, we highlight its relevance in the context of synthesis planning and present a comprehensive overview of how it is built on top of not only our work but also of other recent advancements in the field. We also describe in detail how chemists can easily interact with these capabilities via user-friendly interfaces. ASKCOS has assisted hundreds of medicinal, synthetic, and process chemists in their day-to-day tasks by complementing expert decision making and route ideation. It is our belief that CASP tools are an important part of modern chemistry research and offer ever-increasing utility and accessibility.
- Open Access
- 3 September 2024
- Published by Wiley in Angewandte Chemie
- Vol. 136 (43)
Reproducing Reaction Mechanisms with Machine‐Learning Models Trained on a Large‐Scale Mechanistic Dataset
Mechanistic understanding of organic reactions can facilitate reaction development, impurity prediction, and in principle, reaction discovery. While several machine learning models have sought to address the task of predicting reaction products, their extension to predicting reaction mechanisms has been impeded by the lack of a corresponding mechanistic dataset. In this study, we construct such a dataset by imputing intermediates between experimentally reported reactants and products using expert reaction templates and train several machine learning models on the resulting dataset of 5,184,184 elementary steps. We explore the performance and capabilities of these models, focusing on their ability to predict reaction pathways and recapitulate the roles of catalysts and reagents. Additionally, we demonstrate the potential of mechanistic models in predicting impurities, often overlooked by conventional models. We conclude by evaluating the generalizability of mechanistic models to new reaction types, revealing challenges related to dataset diversity, consecutive predictions, and violations of atom conservation.
- 3 September 2024
- Published by Wiley in Angewandte Chemie
- Vol. 136 (43)
Open AccessReproducing Reaction Mechanisms with Machine‐Learning Models Trained on a Large‐Scale Mechanistic Dataset
Mechanistic understanding of organic reactions can facilitate reaction development, impurity prediction, and in principle, reaction discovery. While several machine learning models have sought to address the task of predicting reaction products, their extension to predicting reaction mechanisms has been impeded by the lack of a corresponding mechanistic dataset. In this study, we construct such a dataset by imputing intermediates between experimentally reported reactants and products using expert reaction templates and train several machine learning models on the resulting dataset of 5,184,184 elementary steps. We explore the performance and capabilities of these models, focusing on their ability to predict reaction pathways and recapitulate the roles of catalysts and reagents. Additionally, we demonstrate the potential of mechanistic models in predicting impurities, often overlooked by conventional models. We conclude by evaluating the generalizability of mechanistic models to new reaction types, revealing challenges related to dataset diversity, consecutive predictions, and violations of atom conservation.
- Open Access
- 2 September 2024
- Published by Wiley in Angewandte Chemie International Edition
- p. e202411296
Reproducing Reaction Mechanisms with Machine‐Learning Models Trained on a Large‐Scale Mechanistic Dataset
Mechanistic understanding of organic reactions can facilitate reaction development, impurity prediction, and in principle, reaction discovery. While several machine learning models have sought to address the task of predicting reaction products, their extension to predicting reaction mechanisms has been impeded by the lack of a corresponding mechanistic dataset. In this study, we construct such a dataset by imputing intermediates between experimentally reported reactants and products using expert reaction templates and train several machine learning models on the resulting dataset of 5,184,184 elementary steps. We explore the performance and capabilities of these models, focusing on their ability to predict reaction pathways and recapitulate the roles of catalysts and reagents. Additionally, we demonstrate the potential of mechanistic models in predicting impurities, often overlooked by conventional models. We conclude by evaluating the generalizability of mechanistic models to new reaction types, revealing challenges related to dataset diversity, consecutive predictions, and violations of atom conservation.
- 2 September 2024
- Published by Wiley in Angewandte Chemie International Edition
- p. e202411296
Open AccessReproducing Reaction Mechanisms with Machine‐Learning Models Trained on a Large‐Scale Mechanistic Dataset
Mechanistic understanding of organic reactions can facilitate reaction development, impurity prediction, and in principle, reaction discovery. While several machine learning models have sought to address the task of predicting reaction products, their extension to predicting reaction mechanisms has been impeded by the lack of a corresponding mechanistic dataset. In this study, we construct such a dataset by imputing intermediates between experimentally reported reactants and products using expert reaction templates and train several machine learning models on the resulting dataset of 5,184,184 elementary steps. We explore the performance and capabilities of these models, focusing on their ability to predict reaction pathways and recapitulate the roles of catalysts and reagents. Additionally, we demonstrate the potential of mechanistic models in predicting impurities, often overlooked by conventional models. We conclude by evaluating the generalizability of mechanistic models to new reaction types, revealing challenges related to dataset diversity, consecutive predictions, and violations of atom conservation.