Abstract:Chemical reactions are the fundamental building blocks of drug design and organic chemistry research. In recent years, there has been a growing need for a large-scale deep-learning framework that can efficiently capture the basic rules of chemical reactions. In this paper, we have proposed a unified framework that addresses both the reaction representation learning and molecule generation tasks, which allows for a more holistic approach. Inspired by the organic chemistry mechanism, we develop a novel pretraining framework that enables us to incorporate inductive biases into the model. Our framework achieves state-of-the-art results on challenging downstream tasks. By possessing chemical knowledge, our generative framework overcome the limitations of current molecule generation models that rely on a small number of reaction templates. In the extensive experiments, our model generates synthesizable drug-like structures of high quality. Overall, our work presents a significant step toward a large-scale deep-learning framework for a variety of reaction-based applications.

What problem does this paper attempt to address?

### Problems Addressed by the Paper This paper aims to address several key issues in chemical reaction representation learning and condition-based molecular generation tasks, proposing a unified deep learning framework to tackle these challenges. #### Main Issues: 1. **Chemical Reaction Representation Learning**: Existing methods for chemical reaction representation overlook fundamental principles of organic chemistry, leading to limited performance. For example, masking bonds or atoms outside the reaction center results in the loss of important information. 2. **Molecular Generation Tasks**: Traditional template methods rely on predefined building blocks and reactions, which restricts the accessible chemical space. Additionally, in drug design, there is a growing need to use chemical reactions as editing tools to modify given structures. #### Solutions: 1. **Self-Supervised Contrastive Learning Framework**: By designing specific self-supervised tasks (such as active center prediction, pairing main reactants with by-products, etc.), this framework can capture the fundamental rules of chemical reactions and embed this knowledge into the model. 2. **Conditional Generation Model**: A template-free conditional generation model is proposed, which can generate multiple synthesizable analogs from a given seed structure, thus overcoming the reliance on limited templates in existing methods during the generation process. 3. **Comprehensive Evaluation**: Through the evaluation of multiple downstream tasks (such as classification, retrieval, and generation), the framework demonstrates superior performance in capturing chemical rules and generating high-quality drug-like structures. In summary, this paper proposes a new unified framework that not only improves chemical reaction representation learning but also excels in molecular generation tasks, showcasing its broad application potential in fields such as drug design.

Bridging the Gap between Chemical Reaction Pretraining and Conditional Molecule Generation with a Unified Model

PrefixMol: Target- and Chemistry-aware Molecule Design Via Prefix Embedding

Chemical-Reaction-Aware Molecule Representation Learning

Learning Chemical Reaction Representation with Reactant-Product Alignment

Unified Deep Learning Model for Multitask Reaction Predictions with Explanation

Prediction of chemical reaction yields with large-scale multi-view pre-training

Bridging Chemical Knowledge and Machine Learning for Performance Prediction of Organic Synthesis.

A Unified View of Deep Learning for Reaction and Retrosynthesis Prediction: Current Status and Future Challenges

Exploring Chemical Reaction Space with Machine Learning Models: Representation and Feature Perspective

Beyond the Typical: Modeling Rare Plausible Patterns in Chemical Reactions by Leveraging Sequential Mixture-of-Experts

Chimera: Accurate retrosynthesis prediction by ensembling models with diverse inductive biases

Self-Supervised Contrastive Molecular Representation Learning with a Chemical Synthesis Knowledge Graph

Hierarchical Framework for Retrosynthesis Prediction with Enhanced Reaction Center Localization

Multi-granularity Score-based Generative Framework Enables Efficient Inverse Design of Complex Organics

Unlocking comprehensive molecular design across all scenarios with large language model and unordered chemical language

A deep learning framework for accurate reaction prediction and its application on high-throughput experimentation data

Bidirectional generation of structure and properties through a single molecular foundation model

PRESTO: Progressive Pretraining Enhances Synthetic Chemistry Outcomes

Multi-constraint molecular generation based on conditional transformer, knowledge distillation and reinforcement learning

BatGPT-Chem: A Foundation Large Model For Retrosynthesis Prediction

Masked Molecule Modeling: A New Paradigm of Molecular Representation Learning for Chemistry Understanding