SCP4ssd: A Serverless Platform for Nucleotide Sequence Synthesis Difficulty Prediction Using an AutoML Model.

Area of Science:

• Synthetic Biology
• Bioinformatics
• Machine Learning

Background:

• DNA synthesis is fundamental for constructing DNA sequences in synthetic biology.
• Sequence features like GC content and repeats influence DNA synthesis difficulty and cost.
• Local sequence characteristics, often overlooked, may also impact synthesis outcomes.

Purpose of the Study:

• To develop a novel automated machine learning (AutoML) approach for predicting DNA synthesis difficulty.
• To identify previously neglected local sequence features affecting DNA synthesis.
• To improve the accuracy and reduce the cost of DNA synthesis through reliable prediction.

Main Methods:

• Implementation of an automated machine learning (AutoML) pipeline.
• Analysis of both global and local sequence features.
• Experimental validation using ten genes from *Escherichia coli* strain MG1655.

Main Results:

• The AutoML model achieved a high F1 score of 0.930, surpassing the current state-of-the-art.
• Identified significant local sequence features that impact DNA synthesis difficulty.
• Experimental validation demonstrated an 80% accuracy, exceeding previous benchmarks.

Conclusions:

• The proposed AutoML approach provides a more accurate prediction of DNA synthesis difficulty.
• Incorporating local sequence features enhances prediction accuracy.
• A cloud-based platform (SCP4SSD) was developed for user accessibility.