When Yield Prediction Does Not Yield Prediction: An Overview of the Current Challenges

Area of Science:

• Chemoinformatics
• Computational Chemistry
• Machine Learning

Background:

• Predicting advanced chemical properties like yield and synthesis feasibility is crucial for chemical research.
• Current machine learning (ML) models face challenges due to the high dimensionality and numerous variables involved in chemical predictions.

Purpose of the Study:

• To systematically evaluate the effectiveness of current ML methodologies in chemoinformatics.
• To identify milestones and limitations of ML in predicting chemical properties.
• To examine data availability and transferability issues through a case study.

Main Methods:

• Systematic review of ML methodologies applied in chemoinformatics.
• Evaluation of ML model performance for predicting chemical properties.
• Case study analysis focusing on data availability and transferability.

Main Results:

• Current ML techniques show promise but face significant hurdles in predicting complex chemical properties.
• High dimensionality and numerous variables (reactants, catalysts, conditions) complicate model development.
• Data availability and transferability remain critical challenges in the field.

Conclusions:

• Reliable ML models can optimize high-throughput experiments and enhance retrosynthetic prediction.
• Further research is needed to address data limitations and improve model generalizability in chemoinformatics.
• Addressing data challenges is key to unlocking the full potential of ML in chemical property prediction.