Logo image
Back
Restriction Synthesis and DNA Restriction Site Analysis Using Machine Learning
Book chapter

Restriction Synthesis and DNA Restriction Site Analysis Using Machine Learning

Ethan Jacob Moyer and Anup Das
Biomedical Sensing and Analysis: Signal Processing in Medicine and Biology, pp 1-53
01 Jan 2022
DOI: https://doi.org/10.1007/978-3-030-99383-2_1

Abstract

Gene synthesis k-fold cross validation Neural network Restriction enzymes Subsequence classification Machine Learning
Restriction synthesis is a novel synthetic deoxyribonucleic acid (DNA) synthesis method. It utilizes iterative restriction enzyme digest and sticky-end alignment to build a target sequence from a reference sequence. In this work, this method is modeled computationally to analyze its efficacy and cost per base pair. By synthesizing many genes from a well-curated generalized reference sequence, we hypothesized that the efficacy and cost of restriction synthesis will greatly improve. In this work, a generalized reference sequence is constructed from classified subsequences and used to synthesize many unique query sequences. We used six different machine learning methods to explore which subsequences to include in this generalized reference sequence: Support Vector Machine (SVM), random forest, naive bayes, K-Nearest neighbor, artificial neural network, and convolutional neural network. Before applying these methods to the data, a series of data curation and feature selection steps create a representative feature space of 231 predictors. In the SVM analysis, unsupervised Principal component analysis was used to reduce the dimensionality of the data. Following these preprocessing steps, six different pipelines classified subsequences based on their nucleotide sequence and other relevant features corresponding to the restriction sites of over 200 restriction enzymes. We used 10-fold cross-validation on a test constant data set to evaluate each model. The random forest model with 45 decision trees performed the best with a classification accuracy of 95.53%, precision of 94.12%, recall of 96.86%, and F1-measure of 95.47%. This model was then applied on unseen data to build the generalized reference sequence. This generalized reference sequence was benchmarked against the randomized reference sequence when executing restriction synthesis simulations. Overall, the results of this work showed an improvement in efficacy and cost of restriction synthesis, supporting the original hypothesis.

Metrics

1 Record Views

Details

Logo image