Performance of neural network basecalling tools for Oxford Nanopore sequencing

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Basecalling, the process of converting raw electrical signals into nucleotide sequences, is essential for Oxford Nanopore Technologies (ONT) sequencing platforms.
• Evaluating basecalling tool performance involves assessing accuracy at both individual read and consensus levels within assemblies.
• Additional factors like taxon-specific training, larger neural network models, and signal-level analysis (Nanopolish) are explored.

Purpose of the Study:

• To compare the performance of different basecalling tools for Oxford Nanopore Technologies sequencing.
• To investigate the impact of taxon-specific training data on basecalling accuracy.
• To assess the benefits of using larger neural network models and signal-level polishing for improved consensus accuracy.

Main Methods:

• Comparative analysis of various basecalling algorithms.
• Evaluation of basecalling accuracy at single-read and consensus levels.
• Assessment of custom model training using taxon-specific datasets and larger neural networks.
• Investigation of Nanopolish for signal-level analysis to improve consensus accuracy.

Main Results:

• Training basecallers on taxon-specific data significantly enhances consensus accuracy, particularly by reducing errors in methylation motifs.
• Larger neural network models improve both read and consensus accuracy but reduce processing speed.
• Nanopolish signal-level analysis can mitigate some basecaller accuracy differences, though initial basecaller accuracy influences post-polish results.

Conclusions:

• Basecalling accuracy has markedly improved recently, with the current ONT Guppy basecaller offering a good balance of accuracy and speed.
• For superior accuracy, users can develop custom models using larger neural networks or species-specific training data.
• Strategic basecaller selection and optimization are key for maximizing accuracy in nanopore sequencing applications.