Abstract:Crystallization is an important process in a broad range of industries, though studies on this topic remain complicated. Recently, machine learning has been applied to resolve complex issues in chemistry and material science. Here we present a machine learning model to propose crystallization experiments for organic small molecules. This model has been integrated into a robotic platform that performs experiments automatically. To improve applicability and accuracy, the model was trained on both simulated and experimental data. In comparative case studies, polymorph screening experiments by the platform yielded a high rate of solid products, and the number of forms obtained by the platform equaled those obtained by human researchers. The model provides a data-based perspective of the promoting and inhibiting influences to crystallization from molecular and interaction features. This work demonstrates the feasibility of applying machine learning techniques to solid-state studies to boost efficiency and deepen understanding.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the complexity and inefficiency in the optimization of crystallization conditions for small organic molecules and polymorph screening. Specifically, the crystallization process plays a crucial role in material discovery and development. Especially in the pharmaceutical industry, the polymorphism of active pharmaceutical ingredients (APIs) is very common, and the selection of an appropriate solid form has a significant impact on drug performance and manufacturability. However, the crystallization process in solution is affected by multiple factors, and it is difficult to quantify the individual contribution of each variable, resulting in traditional polymorph screening experiments being both time - consuming and experience - dependent. To address these issues, the authors developed a machine - learning (ML) - based model for recommending the optimal crystallization conditions for small organic molecules and integrated it into an automated experimental platform. The model overcomes the limitations of existing methods in the following ways: 1. **Data - driven crystallization condition prediction**: Train the machine - learning model by combining simulation data and experimental data to improve the accuracy and applicability of the model. 2. **Automated experimental platform**: Integrate the machine - learning model with a custom - made workstation to achieve automated crystallization experiments, thereby accelerating the experimental process and reducing human error. 3. **Efficient polymorph screening**: Predict crystallization conditions through the machine - learning model, reducing the number of unnecessary experiments and increasing the probability of discovering new polymorphs. ### Main contributions - **Improve experimental efficiency**: Compared with traditional manual experiments, this platform can complete more experiments in a shorter time and has a higher experimental success rate. - **Reveal the influence of molecular features on crystallization**: Through contribution analysis of the model, certain molecular and interaction features that promote or inhibit crystallization were discovered, providing a new perspective for further understanding the crystallization mechanism. - **Expand the application range**: This intelligent automated platform is not only applicable to evaporation and anti - solvent addition crystallization methods but can also be extended to other crystallization methods, having broad application prospects. Through these improvements, this study demonstrates the effectiveness of applying machine - learning and automation technologies to crystallization research, providing new ideas for the construction of intelligent automated systems in the field of materials chemistry.

Machine Learning-Based Recommendation of Optimal Crystallization Conditions for Organic Small Molecules

Highly Efficient Crystallization Studies through Machine Learning and Automation

Machine Learning Methods to Improve Crystallization through the Prediction of Solute–Solvent Interactions

Predicting Natural Rubber Crystallinity by a Novel Machine Learning Algorithm Based on Molecular Dynamics Simulation Data

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

Discovery of Crystallizable Organic Semiconductors with Machine Learning

Machine-Learning Discovered Crystallization Model for Two Dimensional Covalent Organic Frameworks: Towards Precise Control of the Crystal Quality

A Robust Crystal Structure Prediction Method to Support Small Molecule Drug Development with Large Scale Validation and Prospective Studies

Using Machine Learning To Predict Suitable Conditions for Organic Reactions

Machine learning for the structure-energy-property landscapes of molecular crystals

Crystal structure prediction with machine learning-based element substitution

A Hybrid Machine Learning Approach for Structure Stability Prediction in Molecular Co-crystal Screenings

Machine Learning to Instruct Single Crystal Growth by Flux Method

Machine Learning-Based Prediction of Crystal Systems and Space Groups from Inorganic Materials Compositions

Explainable Synthesizability Prediction of Inorganic Crystal Polymorphs using Large Language Models

Taming Two‐Dimensional Polymerization by a Machine‐Learning Discovered Crystallization Model

Machine learning assisted prediction of organic salt structure properties

Towards machine learning for microscopic mechanisms: a formula search for crystal structure stability based on atomic properties

Explainable Synthesizability Prediction of Inorganic Crystal Structures using Large Language Models

Nucleation patterns of polymer crystals analyzed by machine learning models

Machine learning approaches for feature engineering of the crystal structure: Application to the prediction of the formation energy of cubic compounds