Machine learning-aided scoring of synthesis difficulties for designer chromosomes

Area of Science:

• Synthetic biology
• Genomics
• Machine learning applications

Background:

• Designer chromosomes have diverse applications in medicine and biofuels.
• Chromosome fragments can impede artificial chromosome synthesis, limiting technological advancement.

Purpose of the Study:

• To develop an interpretable machine learning framework for predicting designer chromosome synthesis difficulties.
• To identify sequence features contributing to synthesis challenges.

Main Methods:

• An interpretable machine learning framework was developed.
• Six key sequence features influencing synthesis difficulty were identified.
• An eXtreme Gradient Boosting model was trained and validated.

Main Results:

• The model achieved high predictive performance (AUC 0.895 cross-validation, 0.885 test set).
• A synthesis difficulty index (S-index) was proposed for scoring chromosomes.
• Significant variability in synthesis difficulty across different chromosomes was observed.

Conclusions:

• The developed framework accurately predicts and quantifies designer chromosome synthesis difficulties.
• The S-index facilitates the optimization of synthesis processes and genome rewriting.
• This approach has potential applications from prokaryotes to eukaryotes.