Use of a neural network to predict normalized signal strengths from a DNA-sequencing microarray

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Microarray DNA sequencing generates a 4xN data matrix representing signal strengths for each base.
• Variations in signal intensity can reduce the accuracy and confidence of DNA base calls.
• The precise factors influencing these signal variations are not fully understood.

Purpose of the Study:

• To develop and evaluate a neural network model for predicting normalized signal intensities in microarray DNA sequencing.
• To enhance the accuracy and confidence of base calls in DNA sequencing data.
• To investigate the impact of n-gram encoding and network architecture on prediction performance.

Main Methods:

• A feed-forward back-propagation neural network was employed.
• DNA sequences (N=15,453 bases) were encoded using n-gram neural input vectors (n=1, 2, and composite).
• Data was partitioned into training, validation, and testing sets for model evaluation.

Main Results:

• The neural network model achieved overall regression values exceeding 99%.
• Performance improved with an increased number of hidden layer neurons and n-gram composition.
• A very low mean square error was observed, indicating high predictive performance.

Conclusions:

• Neural network modeling effectively predicts normalized signal intensities in microarray DNA sequencing.
• The approach enhances the accuracy of DNA base calling, addressing signal variation challenges.
• Optimizing network architecture and n-gram features is crucial for high-performance DNA sequence analysis.