Transfer Learning Bayesian Optimization to Design Competitor DNA Molecules for Use in Diagnostic Assays.

Area of Science:

• Biomolecular Engineering
• Computational Biology
• Synthetic Biology

Background:

• Engineered biomolecular devices require custom biological sequences, often necessitating extensive and costly laboratory experiments for optimization.
• Developing numerous similar sequences for specific applications presents a significant challenge in terms of time and resources.

Purpose of the Study:

• To present a novel transfer learning design of experiments workflow to reduce the number of experiments required for biological sequence optimization.
• To demonstrate the feasibility of this approach by sharing information between optimization tasks, thereby lowering development costs.

Main Methods:

• Implementation of a transfer learning surrogate model integrated with Bayesian optimization.
• Utilizing cross-validation to assess the predictive accuracy of various transfer learning models.
• Comparison of model performance for both single-objective and penalized optimization tasks.

Main Results:

• Significant reduction in the total number of experiments required for biological sequence development.
• Demonstrated effectiveness using data from the optimization of DNA competitors for diagnostic assays.
• Comparative analysis highlighting the predictive accuracy and performance of different transfer learning models.

Conclusions:

• The proposed transfer learning design of experiments workflow effectively minimizes experimental costs and time in biological sequence development.
• This approach offers a feasible solution for optimizing tailor-made biological sequences, crucial for advancing engineered biomolecular devices.